Harman Research Proves Anechoic Data Predicts Speaker Preference
CIRMMT
Loudspeaker quality depends on flat on-axis response and smooth off-axis directivity rather than electronic room equalization, which degrades sound by trying to correct spatial boundary reflections.
Sean Olive's anechoic Spinorama algorithm achieved an 0.86 correlation with double-blind room listening preference scores across 70 loudspeakers, reaching 0.995 on a subset of 13 models previously misrated by Consumer Reports. (48:00)
Monophonic blind listening tests are far more revealing of spectral defects than stereophonic trials, and speakers that win in mono testing consistently win in stereo without exception. (8:20)
Standard single-microphone in-room calibration systems corrupt audio quality above the transition frequency by attempting to equalize boundary comb filtering and directivity errors that electronic filters cannot physically correct. (35:51)
Loudspeaker price does not correlate with acoustic performance: an $1,800 well-engineered system approached near-flawless anechoic accuracy, while boutique designs costing up to $24,000 exhibited severe resonances and directivity discontinuities. (62:48)
Detailed notes
## Background
This lecture was published on YouTube by CIRMMT (Centre for Interdisciplinary Research in Music Media and Technology) on May 14, 2015. Recorded on April 16, 2015, at Tanna Schulich Hall as part of the CIRMMT Distinguished Lectures in the Science and Technology of Music, the presentation features Dr. Floyd Toole, a consultant to Harman International, discussing the intersection of art, science, opinions, and facts in sound reproduction.
## TL;DR
Acoustic researcher Floyd Toole breaks down decades of scientific inquiry into loudspeaker performance, demonstrating how controlled double-blind listening tests correlate with comprehensive anechoic measurements known as the Spinorama. He exposes common industry fallacies, explaining why sighted listening introduces severe psychological bias, why flat on-axis response alone fails without smooth off-axis dispersion, and why equalizers cannot fix fundamental directivity flaws. Toole introduces the concept of the 'circle of confusion' across music production and playback, showing how rigorous psychoacoustics and standardized multi-angle measurements can accurately predict consumer preference and transcend costly audiophile myths.
## Atmosphere & style
The presentation carries the engaging, authoritative weight of a seasoned scientific pioneer demystifying decades of audio industry folklore with dry humor and sharp wit. Toole speaks in an accessible yet uncompromising academic tone, casually puncturing marketing illusions and flawed calibration practices with robust experimental data. Watching the lecture feels like attending a masterclass in empirical rigor, shifting from wry historical anecdotes to rigorous psychoacoustic charts with conversational ease and infectious clarity.
## Bulletin
- [48:00](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2880s) Sean Olive's anechoic Spinorama algorithm achieved an 0.86 correlation with double-blind room listening preference scores across 70 loudspeakers, reaching 0.995 on a subset of 13 models previously misrated by Consumer Reports.
- [08:20](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=500s) Monophonic blind listening tests are far more revealing of spectral defects than stereophonic trials, and speakers that win in mono testing consistently win in stereo without exception.
- [35:51](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2151s) Standard single-microphone in-room calibration systems corrupt audio quality above the transition frequency by attempting to equalize boundary comb filtering and directivity errors that electronic filters cannot physically correct.
- [01:02:48](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3768s) Loudspeaker price does not correlate with acoustic performance: an $1,800 well-engineered system approached near-flawless anechoic accuracy, while boutique designs costing up to $24,000 exhibited severe resonances and directivity discontinuities.
## Key takeaways
- [06:07](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=367s) Subjective listening evaluations must be conducted as rigorous, level-matched double-blind experiments to prevent brand, price, and visual aesthetics from completely distorting perceived sound quality.
- [08:20](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=500s) Monophonic testing is far more critical and resolving of spectral defects than stereophonic listening, and loudspeakers that win in mono consistently win in stereo.
- [17:00](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1020s) Human auditory perception effortlessly 'listens through' room reflections, mentally decoupling the intrinsic timbre of a loudspeaker from the surrounding acoustic space.
- [22:33](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1353s) No single frequency response curve can describe a loudspeaker; standardized 'Spinorama' measurements capturing 70 anechoic angles across horizontal and vertical orbits are necessary to predict in-room sound.
- [27:53](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1673s) Flat on-axis frequency response alone does not guarantee good sound if off-axis dispersion and sound power curves exhibit irregular discontinuities.
- [30:54](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1854s) Above roughly 300 to 400 Hz, the direct sound and early reflection properties of a loudspeaker dominate in-room response, while room modes and standing waves dictate behavior below that transition.
- [33:44](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2024s) Electronic room equalization cannot fix off-axis dispersion flaws or erratic directivity; smoothing an in-room curve with an EQ destroys the direct on-axis frequency balance.
- [37:51](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2271s) Automated DSP room correction relies on the false assumption that an omnidirectional microphone processes sound fields identically to human binaural hearing and Auditory Scene Analysis.
- [44:35](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2675s) The audio industry suffers from a 'circle of confusion' where unstandardized monitor loudspeakers cause recording engineers to miscalculate equalizations, creating inconsistent playback across consumer environments.
- [53:44](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3224s) Trained listeners with normal hearing exhibit the highest reliability and consistency in evaluations, whereas audio sales and marketing personnel perform the most erratically.
- [55:39](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3339s) A high-fidelity loudspeaker requires three foundational attributes: a flat on-axis response, a smoothly changing directivity index, and complete absence of resonances.
- [58:46](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3526s) Low-frequency reproduction accounts for approximately 30% of overall subjective sound quality ratings, where a 4 to 5 dB shift at 50 Hz doubles perceived loudness due to compressed equal-loudness contours.
- [01:00:03](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3603s) Standard manufacturer loudspeaker specifications are functionally inadequate because a ±3 dB specification represents a wide 6 dB window that conceals audible resonant peaks.
- [01:11:35](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=4295s) Equalizing out micro-acoustic reflection artifacts, such as comb filtering caused by high-backed theater seats at the microphone position, degrades the sound of accurate loudspeakers.
## Detailed walkthrough
### [00:17](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=17s) Art and Opinions vs. Science in Sound Reproduction
Floyd Toole opens his lecture with an urge to move quickly, noting that live sound production is waiting to take the stage later in the program. Addressing a colorful title slide, he points out that art and opinions are deliberately rendered in multiple colors and shifting shades because they represent personal, subjective preference: art and opinions can be whatever anyone wants them to be. Science, by contrast, is stark black and white. It represents an unyielding reality that is either present or absent; if a scientific finding is accurate, it is completely repeatable whether conducted today, tomorrow, locally, or in Timbuktu. Toole stresses that while science and art represent entirely distinct domains, they must collaborate in audio, warning that if they fail to work together, the collective enterprise of entertaining the world through sound reproduction collapses.
Sound production—a live performance unfolding directly on a stage—is art happening in real time, where no two successive performances will ever sound identical. But audio reproduction introduces a fundamental challenge. As Toole explains, “Sound reproduction. You put two letters in front of it. re and the world changes because you're going to try to capture a performance artistically craft it and save it so that somebody somewhere sometime can press a play button and if everything works right they will hear what you've created.” Regrettably, that ideal is almost never realized. What consumers hear in their living rooms, cars, or earbuds rarely matches what the artists and audio engineers originally created in the studio. Toole compares this failure to visual art exhibitions: when an art gallery mounts an exhibition, curators illuminate the paintings with neutral north daylight so viewers see the exact palette the painter applied, rather than bathing the canvases in arbitrary amber or blue spotlights.
Toole also highlights a profound disconnect in acoustic research priorities. Historically, significant funding and academic study have been poured into concert halls, alongside workplace acoustics driven by occupational health regulations. Homes, personal automobiles, and headphones receive far less scholarly attention. Yet consumer listening habits run completely inverse to that research focus: people occasionally attend live concert performances, but the vast majority of human entertainment listening takes place at home, on the road, and through headphones.
- Live sound production is ephemeral, real-time art where no two performances are identical.
- Sound reproduction requires capturing, crafting, and archiving sound for arbitrary future playback.
- Visual art galleries mandate neutral north lighting to preserve an artist's original color palette, whereas audio reproduction routinely colors the source.
- Research historically prioritizes concert halls and occupational spaces over the domestic environments, vehicles, and personal headphones where music is actually consumed.
---
[03:01](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=181s) Slick Marketing, Blind Tests, and the 100-Year Illusion
Turning to the commercial side of audio, Toole examines how manufacturers have manipulated demonstrations for over a century. He presents a historical archival photograph from 1918 featuring Thomas Edison staging a public "tone test." Edison claimed that his newly developed phonograph cylinder playback device could reproduce a human singing voice or a violin so perfectly that listeners could not distinguish the machine from the living artist standing right beside it. Toole draws laughter from the audience by pointing out the glaring flaw visible in the archival image: while the audience members were supposedly participating in a blind comparison, several listeners had their blindfolds slipped down over their ears, turning the event into a blind and partially deaf test.
Beyond Edison's dubious demonstrations, Toole notes that audio companies have repeatedly staged side-by-side comparisons of live musicians playing next to loudspeakers over the past century, routinely dazzling crowds who reported being unable to tell the real instrument from the box. This raises an obvious question: has sound reproduction technology made zero progress since 1918? Toole clarifies that massive technological progress has indeed occurred, but these historic live-versus-recorded events were slick marketing and effective party tricks. In auditoriums and concert halls, the acoustic signature of the venue itself dominates what reaches the audience's ears, heavily masking the loudspeaker's deficiencies. As Toole observes regarding these venue demonstrations, “If they were to do that test in anything other than a really good hall, it wouldn't perform. wouldn't resolve wouldn't work as well.”
- In 1918, Thomas Edison promoted his phonograph through public tone tests claiming perfect parity with live musicians.
- Archival images reveal listeners with blindfolds draped across their ears, rendering the evaluation deaf as well as blind.
- Live-versus-loudspeaker demonstrations rely on dominant hall acoustics to obscure loudspeaker distortion and coloration.
---
[04:20](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=260s) Subjective Measurements and Technical Rigor
To bridge the divide between perception and engineering, Toole emphasizes that listening evaluations must be treated with rigorous scientific discipline. True listening tests are not casual recreational gatherings where people sit around sipping their drink of choice while casually enjoying favorite records; they represent exhausting, disciplined work requiring extensive methodological preparation and sustained mental focus. When designed properly, listening tests yield what Toole designates as "subjective measurements." In these controlled environments, human listeners deliver remarkably consistent, highly repeatable evaluations characterized by standard deviations comparable to precision laboratory instruments.
On the objective side, acoustic measurements must satisfy two mandatory criteria: they must be impeccably accurate, and they must be comprehensive. If data lacks accuracy, it misleads the designer; if it lacks comprehensive coverage across all angles and frequencies, it conceals critical flaws. Toole notes that these strict demands automatically disqualify many audio companies, because they lack the capital and facilities needed to capture such precise acoustic data. Reflecting on his own career, Toole credits his access to world-class infrastructure—specifically the massive anechoic chambers and dedicated listening rooms at the National Research Council (NRC) of Canada and later at Harman International. When comprehensive anechoic measurements are correlated against rigorous subjective listening data, clear patterns immediately emerge. As Toole puts it, “If you do both of those things and start comparing the results, you very, very quickly see patterns, they're not difficult. They're obvious. They just stand up and shout at you.”
```mermaid
flowchart TD
accTitle: Correlation of Subjective and Technical Measurements
accDescr: Diagram illustrating how technical acoustic measurements and subjective listening tests combine to yield repeatable loudspeaker design rules.
subgraph Objective["Technical Measurement Domain"]
A["Anechoic Chamber Measurements"] --> B["Comprehensive 3D Polar Data"]
B --> C["Accurate Technical Baseline"]
end
subgraph Subjective["Subjective Measurement Domain"]
D["Controlled Listening Rooms"] --> E["Double-Blind Experimental Protocols"]
E --> F["Repeatable Listener Ratings"]
end
C --> G["Direct Pattern Correlation"]
F --> G
G --> H["Predictable, High-Preference Loudspeakers"]
```
- Rigorous listening tests demand sustained concentration and yield statistically repeatable subjective measurements.
- Technical measurements must be both completely accurate (to avoid false conclusions) and comprehensive (to capture full off-axis behavior).
- Most audio companies cannot participate in high-level acoustic science because they lack access to large anechoic chambers.
- Research at the National Research Council (NRC) of Canada and Harman proved that loudspeaker design rules directly predicting listener preference become obvious once clean data sets are cross-referenced.
---
[06:07](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=367s) Scientific Listening Test Methodology: Blind, Double-Blind, and Multiple Comparison
Audio discourse is overwhelmed by casual opinions found in enthusiast magazines, consumer websites, and internet discussion forums. Toole dismisses these conflicting opinions as mere background noise; while a handful of posted opinions might accidentally be correct, there is no reliable way to distinguish truth from fantasy without formal scientific methodology. To transform subjective opinions into hard facts, the first mandatory requirement is blinding the evaluation. Listeners must never see the product, because visual awareness of the brand name, retail price, cabinet size, aesthetic finish, or remembered reviews immediately biases auditory perception.
To eliminate experimenter bias, the protocol must be strictly double-blind. If a researcher manually presents the speakers, subtle cues or deliberate ordering can sway listener responses. To prevent this, Toole’s testing protocols have long relied on automated computer software to completely randomize the presentation order while the experimenter leaves the room. Furthermore, playback levels must be rigorously matched, as slight loudness discrepancies distort quality impressions. Toole strongly condemns standard magazine audio reviews where critics take a product home to evaluate it in isolation over weeks. Such home evaluations reveal nothing of scientific value because human hearing readily adapts to linear and non-linear acoustic flaws. True audible differences only emerge through direct, rapid multiple-comparison switching among three or four products simultaneously (A, B, C, and D). In a simple A-versus-B test, if both speakers share a common acoustic flaw, the listener remains blind to it; introducing speaker C and speaker D immediately reveals those shared deficiencies. Toole recalls that he stumbled into this insight through good fortune when his very first listening experiments back in 1966 or 1967 utilized three-way and four-way comparisons.
- Casual audio forum commentary and magazine reviews generate noise that cannot separate truth from psychological bias.
- Blind protocols eliminate visual cues such as brand prestige, retail pricing, cabinet dimensions, and prior review biases.
- Double-blind execution requires automated computer control to remove experimenter expectation and sequencing influence.
- Products must be precisely level-matched, as volume differences directly skew perceived quality.
- Solitary, take-home reviewer auditions teach nothing because human auditory processing quickly acclimates to acoustic errors.
- Multiple-comparison testing (A vs. B vs. C vs. D) exposes shared design flaws that remain hidden during two-product comparisons.
---
[08:20](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=500s) The Power of Monophonic Testing and the Shuffler Mechanism
When conducting loudspeaker comparisons, Toole mandates that the evaluations be conducted monophonically with a single speaker operating in isolation, rather than in stereo or multichannel configurations. Monophonic testing is far more acoustically revealing; listeners become significantly more critical and discerning of spectral defects when evaluating a single sound source. As additional channels are introduced into a room, the spatial complexity distracts the brain, making listeners noticeably less critical of individual loudspeaker flaws. Focusing on monophonic performance also mirrors real-world production demands in television and cinematic film. In cinema sound, the center front loudspeaker carries dialogue and nearly all on-screen Foley effects, making it the most critical transducer in the entire multichannel installation, with high-budget blockbusters being the rare exception that actively pans dialogue. Toole emphasizes that mono excellence directly transfers to stereo: “And every time the the ones that win the mono tests win the stereo tests. There has been no exception to that rule.”
```mermaid
flowchart LR
accTitle: Loudspeaker Shuffler Test Flow
accDescr: Flowchart showing how a listener evaluates randomized, level-matched loudspeakers through an automated shuffler mechanism behind an acoustically transparent screen.
subgraph RoomSetup["Acoustically Transparent Screen"]
A["Shuffled Position C"]
B["Active Reference Position"]
C["Shuffled Position B"]
end
D["Computerized Hydraulic Shuffler"] -->|"3-second physical swap"| B
E["Listener with ABCD Controller"] -->|"Selects Speaker Code"| D
F["Music Loop Playback"] -->|"Automated Randomization"| E
```
Toole details the physical listening room architecture developed at Harman to execute these trials without positional bias. A single listener sits facing an acoustically transparent screen. Using a handheld controller, the subject switches freely between looped musical selections assigned to letter codes A, B, C, or D, while the computer logs their subjective scores and written notes. Behind the acoustically transparent scrim sits a massive, custom-engineered, hydraulically driven, computer-controlled loudspeaker shuffler. Because speaker placement within room boundaries dramatically alters bass and boundary loading, comparisons are invalid if the speakers sit side-by-side. The Harman machine physically moves the chosen loudspeaker into the exact same spatial coordinate in the room within roughly three seconds. Toole acknowledges that building such a pneumatic and hydraulic shuffler cost a fortune, but Harman funded it willingly because removing human bias is paramount: when evaluating sound, human beings are desperately fickle, and their opinions shift radically the instant visual knowledge enters the equation.
- Monophonic evaluation is significantly more critical and revealing than stereophonic or multichannel listening.
- The center channel is the most critical loudspeaker in home theater, reproducing almost all dialogue and on-screen audio.
- Loudspeakers that achieve the highest preference scores in monophonic trials consistently win stereophonic evaluations without exception.
- Comparing speakers placed side-by-side corrupts evaluations due to different boundary interactions and room modes.
- Harman built a custom, hydraulically driven loudspeaker shuffler costing a fortune to physically interchange loudspeakers into the exact same room location within 3 seconds.
- Music loops play continuously while the software randomizes the letter codes assigned to each physical speaker upon each track transition.
[11:12](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=672s) Blind Versus Sighted Listening at Harman
When Floyd Toole arrived at Harman International, he discovered that not a single employee could recall ever participating in a blind test. The prevailing mindset was that they were trained professionals who, by virtue of their experience, were immune to visual bias: “And as a professional, they're not biased by what they see, they really know what good sound is, and so because they know what good sound is, they can do these tests without any complications of blind screening.” Soon after, Sean Olive—Toole’s former colleague from the National Research Council (NRC) and McGill University—joined Harman. Together, they gathered these internal “golden ears” and other listeners to run them through controlled listening tests. A few senior individuals refused to take part, protecting their elite listening reputations because they felt they had everything to lose.
The evaluation compared four loudspeakers. Two were visually identical Harman models selling for around $5,000, differing only in their internal crossover networks. The third was a small, plastic Harman satellite-and-subwoofer system costing $700, which Toole described as “disgustingly good for $700” as long as it was kept at reasonable foreground listening levels without being overdriven. The fourth was a sculptured, wooden-cabinet American competitor priced at $4,000. Under double-blind conditions, listeners identified two clear winners and two inferior systems. However, once the screens were removed and listeners could see the products, the ratings shifted dramatically:
- **Harman employees favored large Harman speakers:** The large, expensive Harman systems received inflated scores simply because listeners recognized the flagship brand identity and impressive enclosures.
- **Prejudice against small and plastic:** The $700 system suffered an immediate penalty in sighted testing; listeners could not overcome the visual assumption that a diminutive plastic box could produce high-fidelity sound.
- **Aesthetic compensation:** The $4,000 competitor earned higher scores when seen than when blind due to its sculpted wood finish, though it still fell short of the high-end Harman speakers.
---
[13:55](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=835s) How Sighted Bias Blinds Listeners to Room and Recording Variations
Room acoustics dramatically alter the acoustic signal reaching the ears, particularly at low frequencies where physical placement dictates standing waves and modal support. In blind evaluations, moving any of the four loudspeakers to different physical positions produced radically different preference scores. Yet when listeners were evaluated sighted, those location-dependent score variations vanished. Listeners concluded that because it was the same physical cabinet, it must sound identical, prompting Toole to observe: “They say, ‘Well, it's the same speaker, it's got to be—has to sound the same. And I gave it a, you know, 5.6 the first time, I'll give it a 5.6 the second time.’ They just kind of go deaf.”
A matching effect occurred with program material. Because commercial recordings possess vastly different spectral balances and instrumentations, different tracks expose unique loudspeaker resonances and flaws. In blind tests, preference curves fluctuated significantly from track to track. In sighted conditions, however, listeners rendered virtually flat, identical scores across tracks, tuning out the audible sonic shifts because their visual perception had already settled on a fixed assessment of the product.
```mermaid
flowchart TD
accTitle: Impact of Blind vs Sighted Evaluations on Auditory Judgment
accDescr: Diagram contrasting how blind testing captures acoustic room and recording variations while sighted testing causes listeners to ignore real acoustic changes due to visual bias.
subgraph BlindCondition["Blind Evaluation (Acoustic Reality)"]
B1["Speaker Position Shifts"] --> B2["Distinctly Different Frequency Response"]
B3["Different Music Tracks"] --> B4["Reveals Specific Flaws / Resonances"]
B2 & B4 --> B5["Scores Fluctuate Strongly Based on Sound"]
end
subgraph SightedCondition["Sighted Evaluation (Cognitive Bias)"]
S1["Visual Recognition of Cabinet / Brand"] --> S2["Preconceptions Formed Before Music Plays"]
S2 --> S3["Location Differences Ignored"]
S2 --> S4["Track Differences Ignored"]
S3 & S4 --> S5["Scores Flatten to Match Visual Expectation"]
end
```
---
[15:26](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=926s) Listener Experience, Demographics, and Audio Mythology
Years of perceptual testing revealed that experienced listeners are substantially more critical and demanding than novices. Untrained listeners typically assign higher average scores across all products because they are easier to please, whereas trained listeners detect subtle colorations immediately. Toole emphasized that gender plays no role in perceptual listening skill: “The listeners of both sexes are just fine, except that more males suffer from hearing loss than females, and so there's the probability that you'll have some faulty opinions from people—from males with hearing loss. We just enjoyed louder hobbies, I guess.”
Visual preconceptions do not just nudge ratings; they establish the perceptual baseline before the audio even starts: “If you believe something, you very likely will hear it.” Toole linked this psychological reality directly to audiophile mythology, pointing out that subjective claims regarding audible differences between exotic speaker cables and similar accessories are purely imagined phenomena born of marketing and sighted bias rather than genuine acoustic transmission differences.
---
[17:00](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1020s) Listening Through Rooms: The Three-Room Experiment
Before leaving Harman, Toole coordinated an experiment (subsequently executed and published) designed to answer whether room acoustics fundamentally invalidate subjective loudspeaker rankings. Three high-performance loudspeakers with very subtle, closely matched differences were tested across three acoustically distinct rooms under strict, double-blind randomized conditions. Despite dramatic physical differences in the rooms' sound fields, listeners ranked the three loudspeakers in virtually the exact same order in every space.
This proved that human hearing possesses the ability to “listen through” an environment rather than conflating the speaker's native sound with the room’s boundary effects. Toole likened this to natural acoustic instruments: one does not need to equalize a Steinway or Bösendorfer concert grand piano specifically for each hall it encounters. Listeners effortlessly decouple the intrinsic timbre of the instrument from the room reflections. For an auditor to rate a loudspeaker reliably, the brain must untangle three competing layers simultaneously:
1. **The loudspeaker’s timbral signature:** Isolating the loudspeaker's direct sound and radiation characteristics.
2. **The room’s acoustic contribution:** Separating room modes, boundary loading, and reflections from the primary source.
3. **The recording’s timbral balance:** Accounting for the unknown equalization and mastering choices made in the studio, despite never having attended the original recording session.
---
[20:11](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1211s) Auditory Scene Analysis and the Need for Multi-Dimensional Measurements
Toole credited McGill’s Albert Bregman and his foundational work on Auditory Scene Analysis for explaining how humans perform this complex mental separation. The human auditory system builds concurrent perceptual streams, processing the sound source and the enclosing space as separate cognitive entities. If an acoustic bass sounds weak in a symphony hall, listeners instinctively blame the hall’s acoustics rather than the double bass player. Similarly, Toole joked that if his speaking voice suddenly took on a boomy, resonant richness, the audience would instantly recognize that he was speaking inside a tiled bathroom rather than assuming his vocal cords had permanently changed: “On a scale of 10, the fidelity rating for my voice is going to be 10. But it's just Floyd in this room as opposed to Floyd in a different room.”
Because human listeners can consistently separate room reflections from a loudspeaker's inherent qualities, researchers can collect stable, repeatable subjective data across diverse programs and listeners. To make this data actionable, Harman set out to build an equivalent database of comprehensive technical measurements to determine which specific engineering metrics correlate with human preference. Toole warned against simplistic solutions, stressing that loudspeaker radiation cannot be captured by a single graph: “We have to admit right away that there is no single curve, no frequency response curve or whatever curve you choose to identify it as, that can completely describe the performance of a loudspeaker.” Because loudspeakers radiate sound omnidirectionally into reflective boundaries, arriving sound fields are far too complex to be deciphered with a single on-axis measurement sweep.
[22:33](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1353s) The Spinorama: Comprehensive Anechoic Measurement vs. In-Room Reversal
Toole addresses the fundamental acoustic challenge of evaluating loudspeakers: you cannot simply place a microphone at a listening position, analyze the complex composite sound field, and mathematically deduce what the loudspeaker originally radiated into the room. As Toole puts it: “You can't go backwards. You can go forwards though.” By measuring the sound radiated in every direction in an anechoic environment, an acoustician can accurately predict the sound that arrives at a listener's ears in an actual room.
This forward-modeling technique relies on the "spinorama" measurement protocol. The procedure entails placing the speaker inside an anechoic chamber and taking 70 precise frequency response measurements at a distance of 2 meters—spinning the cabinet across a full horizontal orbit, laying it on its side, and repeating the process vertically. From this dense sphere of data, several primary components are isolated:
- **Direct Sound (On-Axis):** The baseline first-arrival sound hitting the sweet spot.
- **Listening Window ($\pm 30^\circ$ horizontal, $\pm 10^\circ$ vertical):** A spatial average designed to account for multiple seated listeners rather than a single frozen head position.
- **Early Reflections:** The energetic sum of the first boundary bounces (floor, ceiling, sidewalls, front, and back walls) that represent the loudest secondary arrivals reaching the listening space.
- **Sound Power:** A rigorous, weighted summation of all radiated acoustic energy across every spatial direction indiscriminately.
```mermaid
flowchart TD
accTitle: Spinorama Data Flow
accDescr: Flowchart showing 70 anechoic measurements yielding direct sound, reflections, and sound power curves.
subgraph M["Anechoic Measurements (70 Total at 2m)"]
H["Horizontal Orbit (Spinning)"]
V["Vertical Orbit (On Side)"]
end
subgraph D["Processed Output Curves"]
ON["On-Axis (Direct Sound)"]
LW["Listening Window (±30° H, ±10° V)"]
ER["Early Reflections (Floor, Ceiling, Walls)"]
SP["Total Radiated Sound Power"]
end
H --> D
V --> D
D --> PR["Predicted In-Room Response"]
```
Because the direct sound dictates the primary timbral signature of musical instruments and human voices, the listening window curve must track the on-axis curve as closely as possible. When the early reflections and sound power echo this exact same curve shape, the listener perceives a cohesive, natural acoustic presentation throughout the room.
[25:42](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1542s) Directivity Index, Frequency Scaling, and Industry Reluctance
From the spinorama curves, two Directivity Indexes (DI) are calculated to track the speaker's radiation uniformity across the frequency spectrum. A Directivity Index of zero represents a purely omnidirectional source that sprays equal energy in every direction. This omnidirectional behavior is an unavoidable physical property of subwoofers and woofers at low frequencies, where the transducer is dramatically smaller than the radiated acoustic wavelength:
- At 20 Hz, the acoustic wavelength measures 57 feet.
- At 100 Hz, the acoustic wavelength measures 11.3 feet.
- Compared to these multi-foot wavelengths, a typical one-foot woofer is acoustically tiny, meaning that even at 100 Hz, bass drivers radiate omnidirectionally regardless of which direction they face.
As frequency increases, standard loudspeakers become progressively more directional, beaming energy toward the front until high-frequency radiation narrows significantly. Toole notes that the acoustic ideal is a speaker that maintains smooth, controlled directivity over a wide frequency band, ensuring that room reflections preserve the same timbre as the direct sound.
Although this comprehensive spinorama method was formally standardized in 2013, broad industry adoption faces practical and commercial obstacles. Capturing 70 calibrated anechoic measurements requires anechoic facilities that many manufacturers do not own. Even when third-party testing services are available, commercial transparency presents a hurdle. Toole explains: “And even if they go outside of their own premises and hire a third party to do the measurements, which they could do, they might not like what they see. And so if they don't like what they see, you're not going to see it.”
[27:53](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1673s) The Flat On-Axis Fallacy: The KEF 105.2 Case Study
To demonstrate why conventional single-axis metrics fail, Toole presents data from his research published in 1986 examining a respected, high-end loudspeaker: the KEF 105.2. Its engineering team had executed their strict design brief perfectly by achieving an exceptionally flat, smooth on-axis response. Knowing the designers personally, Toole humorously imagines them celebrating at the local pub after finishing their work. However, the design completely ignored off-axis dispersion. Between $30^\circ$ and $60^\circ$ off-axis, the frequency response became severely disrupted.
When subjected to scientific, double-blind listening evaluations against competing designs, the prestigious loudspeaker neither dominated nor failed completely. Instead, it registered as a distinctly mediocre performer. Evaluating its spinorama curves reveals why:
- The first-arrival sound at the sweet spot was flat and well-behaved.
- The energetic sum of the first four boundary reflections (floor, ceiling, and side walls) exhibited severe response irregularities.
- The total radiated sound power mirrored those off-axis anomalies, completely overriding the smooth on-axis baseline.
The loudspeaker’s audible coloration arose directly from this mismatch between on-axis and off-axis performance. Below approximately 10 kHz, on-axis response ceases to be the dominant factor in room perception. While extreme high frequencies are dominated by direct sound (owing to narrow tweeter dispersion, room absorption, and atmospheric attenuation), every early reflection and boundary bounce below 10 kHz interacts with the listening space. Toole highlights: “The fact that it's beautifully smooth and flat on axis is is not a factor at all.”
[30:54](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1854s) Predicting the In-Room Curve and Managing Room Modes
Summing the temporal sound arrivals—starting with direct sound and progressing through the early and reverberant reflections—produces a steady-state predicted room curve. What a measurement microphone registers inside an actual listening room is fundamentally dominated by the speaker's off-axis dispersion and the acoustic boundaries of the room.
To test whether anechoic spinorama data could accurately predict real-world acoustic behavior, Toole's team placed the loudspeaker into a standard stereo configuration in an actual listening room and averaged measurements across six seated listening positions:
- Moving the speaker a mere 18 inches (half a meter) dramatically shifted its bass response: one position resulted in an overly thick, bloated kick drum, while moving it half a meter away made low-frequency impact nearly vanish.
- Finding an optimal spatial placement yielded balanced bass, producing a tight, articulate kick drum with clear low-end extension.
- Above approximately 300 to 400 Hz, the measured room response across all six seats mapped onto the anechoic spinorama prediction.
Toole explains: “So, there's enough information in the anechoic data of that loudspeaker to be able to predict what's happening in a room above two above three or 400 Hz.” Below 300 Hz, standing waves, eigenmodes, and room resonances corrupt bass performance across small spaces regardless of room shape. Above this transition frequency, what listeners hear is governed by the loudspeaker's off-axis radiated energy interacting with reflective boundaries. Toole concludes that unless someone chooses to listen outdoors, managing reflected off-axis energy is the deciding factor in achieving neutral timbre.
[33:44](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2024s) Why Equalization Cannot Fix Directivity: Predicting In-Room Behavior from Anechoic Data
Floyd Toole begins by pointing out that listening to music outdoors or in an anechoic chamber is far from enjoyable; rooms are inherently beneficial for music, including music reproduced through loudspeakers. However, when engineers or hobbyists take a single in-room measurement, spot a jagged room curve, and instinctively attempt to smooth it out with an equalizer, they end up destroying the only redeeming quality the speaker possessed: its flat on-axis directivity. An equalizer cannot correct off-axis anomalies or directivity mismatches. As Toole puts it plainly: “You can't solve that problem with an equalizer. The only solution for that problem is a loudspeaker that behaves itself.”
```mermaid
flowchart TD
accTitle: Loudspeaker Directivity and Room Curve Prediction
accDescr: Flowchart illustrating how anechoic measurements predict in-room curves and why equalization fails on poorly behaved loudspeakers.
subgraph Anechoic["Anechoic Chamber Measurements"]
OnAxis["On-Axis Frequency Response"]
OffAxis["Off-Axis Frequency Response"]
DI["Directivity Index (DI)"]
SP["Sound Power Curve"]
DI -->|"Inverted"| SP
end
subgraph Room["Room Acoustic Environment"]
RoomCurve["In-Room Sound Field (> 200 Hz)"]
end
subgraph Correction["Intervention Outcome"]
EQ["Parametric / Graphic Equalizer"]
BadOutcome["Alters on-axis response destructively;<br/>fails to fix off-axis radiation"]
end
To demonstrate, Toole presents data from a well-behaved loudspeaker characterized by a flat on-axis response and smooth off-axis dispersion. When calculating the inverse of the directivity index (DI) for a speaker engineered to be flat on axis, that inverted DI curve overlaps the measured sound power curve almost perfectly. Shifting those curves down reveals that they closely predict the actual steady-state room curve measured across a spatial average of six seating positions down to 200 Hz. The only divergence appears above 7 to 8 kHz, where the tweeter’s directional radiation begins to dominate and standard sound power curves underrate its contribution. The central takeaway is that steady-state room curves across the critical audio band are not a mystery: they can be derived directly from comprehensive anechoic polar measurements.
- An inverted directivity index provides an accurate estimate of acoustic sound power when the on-axis response is tuned flat.
- In-room response down to approximately 200 Hz can be predicted directly from anechoic sound power data.
- Above 7 to 8 kHz, sound power curves under-represent the direct contribution of the tweeter, requiring additional directional weighting.
---
[35:52](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2152s) The Illusion of Auto-Room Calibration: Blind Equalizers and Flawed Directivity
Toole shifts to an example of a modern, commercially available loudspeaker that he had listened to earlier that day. Its measurements tell a very different story: the on-axis response is uneven, the off-axis response is erratic, and the directivity index exhibits distinct discontinuities. Because the on-axis response is not flat, the inverted directivity index and sound power curves drift apart, yet the measured sound power still accurately tracks the irregular shape of the room curve. These cold, empirical anechoic graphs are not academic abstractions; they directly explain what a listener hears in a physical room.
A loudspeaker of this type suffers from two independent defects: an uneven on-axis response and compromised directivity. While electronic equalization can easily correct amplitude errors on axis, it is fundamentally incapable of repairing erratic dispersion. When modern consumers run the automated calibration routines found in home audio and home-theater receivers—where pressing a setup button produces test chirps and promises a perfectly tuned system—the processor measures only the combined room curve. Blind to whether a peak or dip originates from room modes, boundary reflections, or driver directivity flaws, the automated routine applies blind filters that often degrade the sound further. Toole stresses: “In this case, equalization is not going to solve the problem. Only a better loudspeaker will.”
---
[37:51](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2271s) The Fallacy of Room Correction Standards and the Dominance of the Loudspeaker
Surveying the audience, Toole finds only a lone hand raised when asking if anyone uses a room equalizer at home or at work. Nonetheless, he notes that room-correction algorithms are ubiquitous in consumer surround-sound receivers, high-end standalone DSP hardware units costing thousands of dollars, and professional calibration services. All of these market solutions rest on the same flawed promise: that placing a microphone in an unknown room playing unknown loudspeakers allows a processor to compute a universal correction. Worse, legacy cinema and broadcast standards promulgated by SMPTE, ISO, and ITU have codified this belief into industry practice.
Toole is uncompromising in his rejection of these legacy standards: “I mean the fact is that when you consider all of the audible characteristics that two ears in the brain can extract from the sound field of sound source in a room, it's not going to be accurately represented by a microphone and a simple analyzer.” A single omnidirectional microphone completely ignores the brain's ability to separate direct sound from early reflections and reverberant decay.
The physical reality is dictated by acoustic scale:
- At low frequencies, room boundaries and standing waves dictate what is heard.
- At higher frequencies, the direct sound and early reflections from the loudspeaker dominate perception.
- The transition frequency where room dominance yields to loudspeaker dominance shifts downward as room volume increases; in an auditorium or performance hall, the loudspeaker is in control down to roughly 40 or 50 Hz.
- Above that transition frequency, installing a properly engineered loudspeaker ensures an excellent listening experience without heavy-handed digital manipulation.
---
[40:27](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2427s) Arthur Benade’s Insight: The Disconnect Between Measurement Microphones and Human Ears
Toole recounts how his philosophy was reinforced by the writings of the late Arthur H. Benade, the renowned physicist and musical acoustician. Before passing away, Benade sent Toole a personal collection of his papers. Benade pointed out a simple truth from a musician’s standpoint: an instrumentalist can play an identical musical riff ten times in succession, producing the exact same acoustic output on stage, and listeners moving around the auditorium will hear that identical tone and timbre every time. Yet, if an acoustic technician moves a measurement microphone between those same audience seats and plots the steady-state transfer function, the graphs vary wildly, displaying dense, erratic comb filtering.
An engineer inspecting those raw concert hall curves in isolation would conclude that the room is an unlistenable acoustic disaster where intelligibility is impossible. Yet human listeners sitting in those seats experience reference-grade acoustic beauty. The human auditory system effortlessly strips away the complex comb-filter interference patterns that confound a test microphone. Toole points out that if one can accurately quantify the radiation properties of the sound source itself—whether a musical instrument or a loudspeaker—one can predict the perceptual experience without trying to force-equalize the chaotic transmission path through the room.
---
[42:18](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2538s) Science in the Service of Art: Closing the Gap Between Recording and Reproduction
Toole draws a direct parallel between the acoustic delivery of live music and electronic sound reproduction. In a live setting, the instrument excites the room, and the listener's brain interprets the combined acoustic field. In sound reproduction, a recording of that instrument must be replayed through a loudspeaker, with the brain performing the same perceptual synthesis. Achieving parity between the two relies entirely on two links in the chain: the technical competence of the loudspeaker designer, and the skill of recording engineers in capturing and storing an uncorrupted acoustic rendering of voices and instruments.
```mermaid
flowchart LR
accTitle: Live vs. Reproduced Sound Chain
accDescr: Parallel between live acoustic generation and the audio reproduction chain from artist to listener.
subgraph Live["Live Acoustic System"]
A1["Musical Instrument / Voice"] --> B1["Acoustic Space / Room"]
B1 --> C1["Human Ears & Brain"]
end
Recording control rooms are packed with massive mixing consoles, outboard racks, and signal processing tools designed to craft an artistic statement. While processing is entirely legitimate for artistic creation, serious problems arise when trying to deliver that art consistently to end consumers across living rooms, automobiles, and headphones. Reflecting on his decades leading acoustic research, Toole summarizes his lifelong operating creed: “Science in the service of art is our business.” While commercial enterprises manufacture steel and wooden enclosures to turn a profit, the ultimate deliverable is neither hardware nor marketing specifications, but the uncompromised experience of good sound.
[44:35](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2675s) The Circle of Confusion and the Dilemma of Inconsistent Monitoring
Loudspeaker sound is the actual product that dictates whether a customer is satisfied, yet determining what constitutes good sound is complicated by what Floyd Toole terms the “circle of confusion.” Loudspeakers are routinely evaluated using commercial recordings, but those very recordings were created in control rooms where recording engineers selected and placed microphones, adjusted equalizers, and added temporal or spatial effects while monitoring the entire process through unstandardized monitor loudspeakers. Over the decades, monitor loudspeakers have varied tremendously. Toole remarks that he has visited modern control rooms equipped with monitor loudspeakers so dreadful that he “wouldn't give house room” to them.
```mermaid
flowchart LR
accTitle: The Circle of Confusion
accDescr: A feedback loop showing how non-standardized monitor speakers corrupt recordings, which are then used to judge consumer speakers.
subgraph Creation["Production Domain"]
A["Recordings Created"] -->|"Monitored through"| B["Control Room Monitors"]
B -->|"Engineer adjusts EQ, mics, levels"| A
end
subgraph Playback["Consumer Domain"]
A -->|"Evaluates"| C["Consumer Loudspeakers"]
C -->|"Listeners judge"| D["Artistic Quality / Sound"]
end
D -.->|"Confusion: Blame speaker, room, or mix?"| Creation
```
Because of this lack of standardization, recordings exhibit vast differences in spectral balance. Toole recounts putting on headphones to test whether streaming music services kept playback volume levels constant between tracks. While level matching was handled reasonably well, the timbral balance swung wildly from track to track, forcing him to adjust basic bass and treble tone controls to make many recordings acceptable. If a monitor speaker in a studio overemphasizes bass, the recording engineer cuts bass from the mix, producing a thin recording on an accurate consumer system. Without a defined relationship between the domain where art is created and the domain where it is appreciated, the audience cannot tell whether an unpleasant sound originates from the recording, the room acoustics, or the loudspeakers themselves.
- “And so, how do we know what good is? We listen. And when we listen, we get caught into what I call the circle of confusion where loudspeakers are evaluated using recordings.” — Floyd Toole
- “There has these professional monitor loudspeakers in rooms have to behave in some ways similar to our consumer loudspeakers and rooms or cars even or the system doesn't work.” — Floyd Toole
- “Do I blame my loudspeakers? Is it my room that's the problem or was it just a crappy recording? That's that's the dilemma you're in. Unless you have measurements.” — Floyd Toole
---
[48:00](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2880s) The Harman CEO’s Challenge and Predicting Listener Preference
The question of whether anechoic Spinorama measurements can predict perceived sound in real rooms reached an urgent turning point following a phone call from the CEO of Harman International. The CEO challenged Toole over a recent *Consumer Reports* issue in which Harman products received poor ratings, pointing out that Toole was paid a substantial salary specifically to ensure Harman products sounded good. When Toole explained that the loudspeaker industry had known for decades that *Consumer Reports* evaluated loudspeakers incorrectly, the CEO told him to take the necessary time and budget to show how to evaluate them properly.
Toole and Sean Olive tackled the problem by pairing psychoacoustics with Harman’s extensive double-blind listening database. Harman routinely tested its own speakers and competitors' models in single-listener, double-blind trials, assigning each a preference score on a 10-point scale based on room playback. At that time, their database contained comprehensive data on 70 loudspeakers. Olive developed an algorithm to analyze anechoic Spinorama data—accounting for on-axis and off-axis spectral balance, bandwidth, and narrow-band resonances—to calculate a predicted preference score. When the calculated scores from pure anechoic data were compared against the actual blind listening scores gathered inside a room, the statistical correlation reached 0.86.
- “And so, one day I get a phone call from the CEO of Harman saying, 'Toole, we pay you a lot of money to make sure our products sound good.'” — Floyd Toole
- “And we have the number that came out of the listening test and the correlation was 0.86. That that's not shabby. You go to Las Vegas with those odds, they'll kick you out of the casino in no time.” — Floyd Toole
---
[50:45](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3045s) Consumer Reports Debunked and the $10,000 Speaker Reality Check
To directly confront the *Consumer Reports* findings, Harman isolated the 13 specific loudspeaker models reviewed in the magazine from their 70-speaker pool. In a side-by-side plot of the double-blind listening results versus the published *Consumer Reports* ratings, the relationship was slightly negative with high scatter: the loudspeaker rated highest in double-blind testing had been ranked near the very bottom by *Consumer Reports*. Conversely, when Olive’s calculated predictive model was applied to those same 13 speakers, the correlation between the calculated prediction and the real blind listening scores was 0.995—representing an almost perfect agreement.
Toole illustrates the practical power of these measurements using four audiophile-approved, high-end loudspeakers costing roughly $10,000 per pair. Magazine reviews gave consumers the impression that any of the four would represent the state-of-the-art. Yet in Harman’s blind tests, two speakers were locked in a close race for first place while the other two lagged far behind. The differences correlated directly with the Spinorama measurements: the flatter and smoother the response, the higher the listening score. Toole emphasizes that no designer would ever tolerate an electronic component that was not flat from 10 Hz to 100 kHz, making it absurd to abandon that principle when evaluating loudspeakers.
- “The square, the open squares are the Consumer Reports ratings. And you can see our best is their worst.” — Floyd Toole
- “When we put it through the algorithm calculating algorithm that Sean had created, the correlation was 0.995. Well that's one, that's perfect. The calculated subjective ratings and the real subjective ratings just agreed.” — Floyd Toole
- “We would never ever buy an electronic component and nobody would ever design an electronic component that didn't have a frequency response that was flat from, you know, 10 hertz to 100 kHz, you know, that kind of thing. It's flat and smooth. It's not not difficult to do and everybody does it. Why would we think that when you get to a loudspeaker, we can kind of throw that rule away?” — Floyd Toole
---
[53:44](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3224s) Listener Reliability and the F-Statistic
The dataset was compiled through extensive listening trials involving 350 participants over three months (with 268 included in Sean Olive’s published Audio Engineering Society paper). The study evaluated how listener background and training influenced reliability, quantified using the F-statistic. This metric evaluated two critical capabilities: consistency (repeating identical ratings for the same loudspeaker when heard multiple times in randomized, blinded sequences) and differentiation (the willingness to assign broad rating spreads, such as giving poor products a 2 or 3 and good products a 6 or 7, rather than clustering everything safely around 5 and 6).
The data revealed sharp contrasts between different listener groups. The most reliable evaluators were trained listeners who had been pre-screened for normal hearing and coached to identify specific spectral and resonance defects. Retail audio salespeople performed quite well, while six audio reviewers who volunteered for the tests delivered middling, unimpressive results. The worst performance came from Harman’s internal sales and marketing staff, who exhibited very poor consistency and differentiation despite representing the products commercially.
- “These are the people are going out into the field say buy our product. It's great. I you can I'll vote for it. Please don't.” — Floyd Toole
- Listener groups ranked by evaluation reliability (highest F-statistic to lowest):
1. Selected and trained listeners (screened for normal hearing and trained in acoustic defect detection)
2. Retail audio salespeople
3. Professional audio reviewers (six volunteers)
4. Harman brand, sales, and marketing personnel
[55:39](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3339s) Humans as Measuring Instruments and the Recipe for Good Sound
Floyd Toole reflects on listening tests conducted with generic, non-audio, non-music students. Although these listeners differed in their consistency and the numerical spread of their scores, their overall rankings of products were completely identical, with zero inconsistencies. Toole expresses continuing awe after 40 years in the field: listeners can sit in an unfamiliar room, listen to unfamiliar music through unfamiliar loudspeakers, and still arrive at highly accurate, shared conclusions about sound quality—provided the test is double-blind. Referring back to Arthur Benade’s ideas, Toole emphasizes that while musical instruments radiate sound into rooms in wildly complex ways that cannot be standardized, loudspeakers reproducing music can and must be measured systematically.
```mermaid
flowchart TD
accTitle: Loudspeaker Design Requirements
accDescr: Flowchart showing the three foundational acoustic requirements for accurate loudspeaker sound reproduction.
subgraph Requirements["Acoustic Performance Goals"]
A["Smooth & Flat On-Axis Response"]
B["Smoothly Changing Directivity Index"]
C["Total Elimination of Resonances"]
end
A --> D["Accurate Timbre & High Listener Preference"]
B --> D
C --> D
```
Combining anechoic measurements with subjective tests reveals clear psychoacoustic relationships. To produce a loudspeaker that listeners consistently praise, three core design rules must be met:
1. The on-axis frequency response must be smooth and flat, or listeners will actively complain.
2. The directivity index must be smooth and relatively constant, or change only gradually across frequency.
3. Resonances must be completely eliminated.
Toole explains that resonances are particularly damaging because human voices and musical instruments are themselves composed of natural acoustic resonances. When a loudspeaker introduces its own mechanical or electrical resonances, it colors every single voice, instrument, and recording played through it. Detecting these flaws requires high-resolution measurement rather than outdated tools; as Toole puts it:
> “The old idea that third-octave data is useful—I'm afraid if you have a third-octave analyzer, give it to your kids because it's a toy.”
---
[58:46](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3526s) The Disproportionate Importance of Bass Performance
Toole turns to low-frequency reproduction, revealing that bass performance accounts for roughly 30% of the overall subjective rating of a loudspeaker. Getting bass right is uniquely challenging due to human hearing physiology:
- In the midrange (roughly 200–300 Hz to 3–4 kHz), perceived subjective loudness doubles or halves with an increment of approximately 10 dB.
- At low frequencies (such as 50 Hz), equal-loudness contours compress significantly. As a result, a shift of merely 4 to 5 dB can produce a two-to-one difference in perceived loudness.
- Because low-frequency loudness changes so drastically over small amplitude shifts, even a minor bass irregularity degrades the perception of the entire frequency spectrum.
---
[01:00:03](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3603s) The Fallacy of Manufacturer Specs and the Power of the Spinorama
Toole shifts to the consumer market, asking whether any of these critical traits can be predicted from standard manufacturer specifications. When shopping, consumers are almost never shown Spinorama data—aside from a few Harman models and products like the JBL M2. Instead, manufacturers typically cite basic frequency response envelopes, such as "58 Hz to 22 kHz, ±3 dB."
A ±3 dB specification defines a massive 6 dB tolerance window. Considering that an increase of 6 dB approaches a doubling of loudness in human perception, this single number obscures enormous potential flaws. A raw frequency response curve provides much more insight than a basic text spec, revealing broad tonal tilts that could easily be adjusted with simple tone controls. However, a full Spinorama provides definitive insight because loudspeaker transducers operate as minimum-phase devices:
- If amplitude response curves show smooth behavior with no peaks across all spatial angles, there are no ringing resonances.
- If a narrow bump appears consistently across all spatially averaged curves, it is an audible resonance that will ring in time.
- Because minimum-phase anomalies can be perfectly eliminated by equalizers matching those bumps, Toole argues that active loudspeakers equipped with onboard digital electronics and amplification represent the true future of high-performance audio design.
---
[01:02:48](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3768s) Real-World Comparisons: From $200 Bargains to $24,000 Flaws
To illustrate how price correlates poorly with actual acoustic quality, Toole walks through real Spinorama measurements across several price tiers:
| Price Point | Speaker Type / Description | Measurement Findings | Audible Consequences |
|:--- |:--- |:--- |:--- |
| **$200 / pair** (on sale for $150) | Budget big-box store bookshelf speaker | Tilted curve with several persistent resonant peaks across spatial averages. | Colorations from ringing resonances; requires EQ to tame. |
| **$1,800 / pair** | High-performance active/passive design | Exceeds its modest "±3 dB" spec; exceptionally flat and resonance-free. | Near perfection in timbral accuracy; classic example of the point of diminishing returns. |
| **$11,000 / pair** | Large electrostatic panel over a conventional box woofer | Severe directivity mismatch; large gap between on-axis and listening window curves; significant ringing peaks. | High directivity means listeners sitting on the same sofa hear different balances; flawed acoustic execution despite "fast mass-less diaphragm" marketing claims. |
| **$24,000 / pair** | Ultra-high-end luxury floorstander | Occupies the entire ±3 dB tolerance; distinct directivity kink; prominent un-equalizable resonance. | Clear, audible coloration well above the threshold of audibility that performs worse than the $1,800 pair. |
| **$16,000 / pair** | High-end monitor / loudspeaker | Audacious manufacturer specification of ±1 dB; barely squeaks inside the limits. | Extremely tight tolerance, pushing the technical limits of passive manufacturing. |
Evaluating the $1,800 pair, Toole notes that standard marketing practices frequently under-represent superior engineering:
> “So for $1,800 a pair, if you were fortunate enough to go out and find those and buy them, you ended up at home with something that is dangerously close to the best that can be done. Period.”
By contrast, the $11,000 electrostatic hybrid relies on romanticized audio lore: audiophiles assume that a diaphragm with virtually no mass must move instantaneously, creating fast and accurate sound. Instead, the measurements prove it rings heavily and suffers from severe directivity problems, leading Toole to declare it “truly a bad design” and an exceptionally poor investment. Similarly, the $24,000 speaker reveals audible resonances and erratic radiation narrowing that cannot be salvaged by equalization, proving that spending more money often yields inferior timbral fidelity.
[01:06:05](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3965s) Production Consistency and the Uselessness of Spec Sheets
Designing an exemplary loudspeaker is only half the battle; reproducing that benchmark reliably on an assembly line is where the industry often stumbles. Floyd Toole points out that building a single "golden prototype" in a laboratory can happen through skill or sheer luck, but cloning that exact performance thousands of times in mass production requires rigorous, expensive engineering. When Harman guarantees that a speaker will perform within strict specifications, customers are paying directly for that manufacturing quality control. Smaller companies frequently lack the resources or infrastructure for strict production tolerances, leaving consumers to gamble on whether the unit they buy matches the unit that was reviewed.
Toole vents his frustration at the audio industry's refusal to supply consumers or audio professionals with meaningful performance data. Looking at marketing materials and specification sheets across both consumer and professional studio monitor websites reveals a glaring absence of standardized, objective metrics. As Toole quips:
> “I and just my joke, there's more useful and reliable information on the side of a tire than there is in a loudspeaker specification sheet.”
Rather than publishing comprehensive measurements of directivity and off-axis behavior, manufacturers simply ask customers to take their claims on faith: “Trust me. Trust us.”
- **The Golden Prototype Dilemma**: Handcrafted prototypes frequently deliver reference-grade sound, but mass production without strict quality control leads to massive sample-to-sample variation.
- **Consumer vs. Professional Specs**: Neither consumer Hi-Fi nor pro-audio monitor spec sheets provide actionable, standardized engineering data to evaluate how a speaker will perform in a room.
---
[01:07:44](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=4064s) The Circle of Confusion and the Clear Window Analogy
Pointing to the JBL M2 Master Reference Monitor, Toole notes that it represents arguably the pinnacle of modern loudspeaker design. Yet, when listeners ask how such an accurate speaker actually sounds, Toole's blunt answer is that it completely depends on what recording is fed into it. Even if engineers manage to build an absolute, mathematically flawless loudspeaker, it cannot make every record sound pleasing because commercial recordings are plagued by immense variability. Because albums are mixed and mastered under completely arbitrary acoustic circumstances—using unmeasured monitors in uncalibrated rooms—listeners cannot know whether to praise or blame the speaker or the recording.
Toole recalls a discussion he had earlier that day in the office of McGill professor Wieslaw Woszczyk, using the metaphor of looking through a window. The recording engineer and artist are responsible for painting the landscape outside, while the monitor loudspeaker acts as the glass window through which they view it:
> “And if I have a blue window and you'll have you, you know, it's it's a silly analogy, but no, it's not a silly analogy because it's really it's real.”
If the window is completely neutral and clear, the artist sees exactly what exists outside; if they dislike the scenery, it is their job as the artist to alter the scene. But if the monitoring window is tinted pink, and the listener at home views it through a window tinted blue, the artistic intent is hopelessly distorted. The only way to break out of this circle of confusion is to enforce neutral loudspeakers on both ends: in the production control room and in domestic playback environments. Fortunately, good acoustic design is becoming far more democratic; today, even budget consumer setups around $200 (paired with a subwoofer), modern car audio systems, and well-engineered portable Bluetooth speakers sound vastly better than equipment did ten or twenty years ago.
```mermaid
flowchart LR
accTitle: The Circle of Confusion Window Analogy
accDescr: Diagram contrasting an uncalibrated monitoring chain with tinted windows versus a neutral monitoring chain with clear windows.
subgraph TintedChain["Distorted Chain (Circle of Confusion)"]
direction LR
A["Artistic Creation / Mix"] --> B["Pink Window<br>(Colored Studio Monitor)"]
B --> C["Blue Window<br>(Colored Home Speaker)"]
C --> D["Unpredictable Result"]
end
subgraph NeutralChain["Calibrated Neutral Chain"]
direction LR
E["Artistic Creation / Mix"] --> F["Clear Window<br>(Accurate Studio Monitor)"]
F --> G["Clear Window<br>(Accurate Home Speaker)"]
G --> H["Intended Artistic Perception"]
end
```
- **The Recording as the Weak Link**: When played through an objectively neutral speaker, an unflattering presentation points directly to flaws baked into the mix, not a defect in the playback monitor.
- **Broadening Access to Neutrality**: Science-based spinorama measurements and target standards have elevated playback quality down to entry-level monitors, consumer Bluetooth speakers, and automotive sound systems.
---
[01:11:35](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=4295s) In-Room Reality: The JBL M2 from Home Theaters to Dubbing Stages
Before concluding, Toole brings up one final slide that illustrates how an accurately designed loudspeaker behaves in radically different acoustic spaces. The slide plots the measured in-room frequency response of the JBL M2 in a modest, six-seat domestic home theater at Harman alongside measurements taken in commercial film sound dubbing stages ranging across 24, 60, 100, 114, and 516 seats. Remarkably, above approximately 200 Hz—the point where the direct sound and directivity of the loudspeaker dominate over room boundary reflections—the frequency response curves across all these drastically different venues track each other almost identically.
Below 200 Hz, room modes and boundary loading take over, presenting a distinct low-frequency challenge that requires multi-subwoofer optimization and targeted equalization. Toole highlights a critical trap visible in the high-frequency measurement curves from the cinema stages: slight irregular ripples appear in traces where technicians placed test microphones directly at seated ear level. Placing an omnidirectional microphone inches away from the reflective boundary of a high-backed theater seat produces severe acoustic comb filtering. While human listeners with two ears and binaural processing naturally ignore these localized seat reflections, a measurement microphone registers them as sharp acoustic peaks and dips. Technicians frequently fall into the trap of aggressively equalizing these microphone artifacts, degrading what was otherwise a pristine loudspeaker response.
> “And in equalizing it, they're equalizing something that cannot should not be equalized and they're making what might have been a really good loudspeaker sound worse.”
- **The ~200 Hz Transition**: Above roughly 200 Hz, an accurate loudspeaker with smooth off-axis directivity produces a consistent, predictable in-room response across both tiny living rooms and 500-seat auditoriums.
- **The Seat-Back Reflection Trap**: Placing calibration microphones directly at ear height next to seat backs registers interference ripples that the human brain ignores, leading sound technicians to ruin good loudspeakers through misguided high-frequency equalization.
## Notable quotes
> [15:26](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=926s) If you believe something, you very likely will hear it. — Floyd Toole
> [01:06:05](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3965s) there's more useful and reliable information on the side of a tire than there is in a loudspeaker specification sheet. — Floyd Toole
> [55:39](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3339s) The old idea that third-octave data is useful—I'm afraid if you have a third-octave analyzer, give it to your kids because it's a toy. — Floyd Toole
> [20:11](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1211s) We have to admit right away that there is no single curve, no frequency response curve or whatever curve you choose to identify it as, that can completely describe the performance of a loudspeaker. — Floyd Toole
> [42:18](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2538s) Science in the service of art is our business. — Floyd Toole
## Mentioned
Floyd Toole, Thomas Edison, National Research Council (NRC) of Canada, Harman International, Sean Olive, McGill University, Albert Bregman, Steinway, Bösendorfer, KEF 105.2, Arthur H. Benade, JBL M2 Master Reference Monitor, Consumer Reports, Audio Engineering Society (AES), SMPTE, ISO, ITU, Wieslaw Woszczyk
## Video description
To view a visualization of the videos recorded by CIRMMT, visit https://www.cirmmt.org/en/videos
CIRMMT Distinguished Lectures in the Science and Technology of Music
Floyd Toole, consultant to Harman International, USA
16 April 2015 - Tanna Schulich Hall
http://www.cirmmt.org/activities/distinguished-lectures
https://www.facebook.com/CIRMMT/
APA video citation:
Toole, F. (2015, May 14). Sound reproduction - art and science/opinions and facts -
CIRMMT Distinguished Lectures in the Science and Technology of Music. [Video file].
Retrieved from https://www.youtube.com/watch?v=zrpUDuUtxPM
Acoustician Floyd Toole argues that sound reproduction must be grounded in objective science rather than subjective marketing, demonstrating that comprehensive anechoic measurements accurately predict human listening preferences across rooms. Achieving true fidelity, however, requires standardizing monitor loudspeakers on both sides of the recording and playback chain to break the industry's pervasive 'circle of confusion.'
Controlled Methodology Versus Audio Sighted Bias (6:07)
Toole contends that casual audio evaluations and commercial demonstrations are distorted by visual bias, marketing cues, and venue acoustics. In double-blind evaluations where listeners cannot see brand badges, finishes, or prices, human subjects deliver remarkably consistent subjective measurements with standard deviations comparable to laboratory instruments. In contrast, sighted testing causes listeners to favor expensive or familiar brands, penalize competent budget enclosures, and ignore genuine acoustic variations introduced by different speaker positions and program material.
To yield scientifically valid data, Toole mandates level-matched, double-blind multiple comparisons (switching rapidly among three or four options) using randomized computer controls. Furthermore, evaluations must be performed monophonically with a single speaker. Because stereo and multichannel setups introduce spatial complexity that distracts the brain from spectral defects, single-speaker testing proves far more revealing, and speakers that win monophonic trials consistently win in stereo testing.
Spinorama Data and In-Room Predictability (22:33)
Because human hearing naturally separates direct sound from early boundary reflections, a single on-axis measurement sweep cannot adequately characterize a loudspeaker. Toole details the standardized 'spinorama' method, which compiles 70 calibrated anechoic measurements across horizontal and vertical orbits at 2 meters. This comprehensive sphere captures the direct sound, the listening window average (±30° horizontal, ±10° vertical), early reflections, and total acoustic sound power, along with directivity indexes.
Research led by Sean Olive demonstrated that pure anechoic spinorama metrics correlate with real-world blind listening scores at 0.86 across 70 loudspeakers, and reached 0.995 on a 13-speaker subset that Consumer Reports had evaluated with near-zero or negative correlation. Anechoic data reliably predicts in-room behavior above a transition frequency of roughly 200 to 400 Hz. Below this transition, room boundaries and standing waves dictate bass response, accounting for roughly 30% of subjective preference ratings due to human equal-loudness sensitivity.
The Limits of Equalization and Directivity Failures (33:44)
Toole demonstrates that high price tags do not guarantee acoustic competence. A well-engineered $1,800 active/passive pair can deliver near-flawless measurements, whereas certain $11,000 electrostatic panels or $24,000 luxury floorstanders suffer from directivity mismatches and prominent resonances. Standard manufacturer specifications, such as ±3 dB envelopes, define a 6 dB tolerance window that frequently conceals audible, ringing resonances.
While electronic equalization can easily correct on-axis amplitude errors, Toole stresses that it cannot repair erratic off-axis dispersion. Automated room calibration software and legacy cinema standards often rely on a single omnidirectional test microphone that registers room comb filtering and seat-back reflections that human binaural hearing naturally ignores. Attempting to force-equalize combined steady-state room curves destroys flat direct sound, degrading well-designed transducers instead of improving them.
The Circle of Confusion in Sound Reproduction (44:35)
Even mathematically accurate loudspeakers cannot guarantee a pleasing experience due to the 'circle of confusion.' In recording studios, mixing engineers choose microphones, adjust equalizers, and balance tracks while monitoring through non-standardized loudspeakers in uncalibrated rooms. If a control room monitor suffers from excessive bass or midrange colorations, the engineer alters the mix to compensate, embedding an inverted defect directly into the commercial recording.
Toole likens monitoring to looking through a window: if the recording studio views the art through a pink-tinted window, and a home consumer listens through a blue-tinted window, the artistic intent is irrecoverably compromised. The only systemic remedy is enforcing objective acoustic standards across both professional production environments and consumer playback equipment.
Toole concludes that neutral sound reproduction requires smooth on-axis response, controlled directivity, and the elimination of mechanical resonances. However, as long as manufacturers conceal directivity metrics behind vague marketing specifications and studios fail to standardize monitor environments, consumers remain trapped in an unstandardized loop where speaker quality cannot be separated from recording quality.
## Background
该视频由 CIRMMT（音乐媒体与技术跨学科研究中心）于 2015 年 5 月 14 日发布在 YouTube 平台上，记录了哈曼国际顾问弗洛伊德·图尔（Floyd Toole）博士于 2015 年 4 月 16 日在麦吉尔大学 Tanna Schulich 音乐厅发表的杰出演讲。演讲主题为“声音重放——艺术与科学／观点与事实”，探讨了声学测量、心理声学评估以及打破音响行业主观认知偏差的科学方法。
## TL;DR
Floyd Toole 深入剖析了声音重放领域中艺术、个人主观臆断与客观科学之间的鸿沟，尖锐批评了音频产业长期依赖营销噱头与非标准化监听的“混乱之环”。他通过数十年严密设计的双盲听音实验证明，人类在排除视觉等心理偏见后展现出惊人一致的听音偏好，平滑的轴向频响与优秀的指向性指数直接决定了音质优劣。依托消声室的 Spinorama（空间自旋）全方位测量数据，工程师能够精准预测扬声器在真实房间中 300-400 Hz 以上的中高频表现，破除单麦克风粗暴室内 EQ 的迷信。演讲以扎实的对比数据揭穿了昂贵发烧器材的营销神话，强调只有录音与回放两端均采用中性扬声器，科学才能真正为艺术服务。
## Atmosphere & style
现场弥漫着资深科学家特有的从容、犀利与幽默，Toole 既像一位经验老到的学界泰斗在拆解物理法则，又像一位愤世嫉俗的行业老兵在无情撕碎商业谎言。他谈吐风趣且底气十足，信手拈来消声室图表与双盲测试统计，在调侃爱迪生营销戏法与高价发烧玄学时金句频出、引得全场会心微笑。整场演讲不仅有严密的逻辑推导，更洋溢着对消除行业乱象的紧迫感与崇高使命感，带给听众一种大浪淘沙后拨云见日般的理性震撼。
## Bulletin
- [08:20](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=500s) 双盲实验证明人类具有极高的听音评估一致性，视觉偏见与品牌暗示会彻底颠覆听觉感知；且单声道盲听能最大程度暴露扬声器缺陷，在单声道测试中胜出的音箱在立体声测试中必然全胜。
- [48:02](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2882s) 哈曼通过消声室 Spinorama 空间多维测量与算法构建的预测模型，客观预测打分与真实双盲听感评分的相关系数高达 0.86（与特定对比组甚至达到 0.995），彻底打破了主观评价与客观参数脱节的迷信。
- [44:35](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2675s) 声音回放领域深陷“混乱之环”，录音端与消费端均缺乏统一的监听标准；唯有两端共同采用轴向平平、指向性顺滑的“中性”扬声器，才能真正还原艺术创作。
- [37:44](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2264s) 传统房间全自动 EQ 存在重大盲区，单麦克风测量无法模拟人类双耳与大脑分离直达声与反射声的听觉机制；盲目拉平中高频曲线往往会破坏音箱原本平直的轴向响应，使好声音恶化。
## Key takeaways
- [00:17](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=17s) 声音重放的本质是科学为艺术服务，其工程目标是精准再现创作者意图，如同美术馆使用中性北向自然光还原画作本色。
- [02:26](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=146s) 百年来让现场乐手与扬声器同台对比的所谓“盲听”大多是营销戏法，大厅混响往往掩盖了器材瑕疵，脱离严密实验的测试毫无科学价值。
- [06:08](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=368s) 双盲盲测是排除品牌、外观与价格等心理偏见的唯一途径；测试必须采用精确声级匹配与即时多路（A/B/C/D）对比。
- [08:20](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=500s) 单声道测试的辨别灵敏度显著高于立体声或多声道，因为声道叠加会掩盖缺陷；单声道盲测中获胜的扬声器在立体声测试中毫无例外地胜出。
- [16:59](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1019s) 人脑具备听觉场景分析能力，能够“穿透房间倾听”，将扬声器音色、房间反射声与录音本身的风格解构剥离。
- [22:33](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1353s) 没有任何单一频响曲线能代表音箱全貌，必须依托消声室 Spinorama 测量全方位记录轴向、聆听窗口、早期反射与声功率。
- [25:42](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1542s) 宽频段内保持平滑且相对恒定的指向性指数（DI）至关重要，只有这样才能确保离轴反射声与直达声的音色匹配统一。
- [30:54](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1854s) 在 300-400 Hz 过渡频率以上，消声室测量数据能够高度吻合并准确预测扬声器在真实房间内的中高频稳态响应。
- [33:44](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2024s) 均衡器（EQ）绝对修不好指向性糟糕的音箱；粗暴拉平麦克风在房间中测得的中高频坑洼，只会彻底毁掉原本平直的轴向直达声。
- [37:51](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2271s) 传统自动房间校准误区在于测试麦克风无法识别人脑感知机制，把不同时间到达的直达声与空间反射声简单线性相加。
- [44:35](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2675s) 音响行业深陷“混乱之环”：录音师依靠非标准的监听设备做混音决策，消费者又用参差不齐的唱片去评价家用回放系统。
- [48:02](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2882s) 哈曼通过 70 款音箱的双盲测试与 Spinorama 算法建模，证实了平直轴向响应与顺滑离轴表现能够极其精准地预测主观偏好打分。
- [52:07](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3127s) 在主观听力测试中，受过批判性训练的听众打分一致性最高且区分度最强，而音响媒体测评人和营销人员在盲听下的辨别能力往往表现极差。
- [58:46](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3526s) 低频响应占据主观品质评价约 30% 的权重；受等响曲线影响，低频 4-5 dB 的偏差即可造成主观响度翻倍，调校失衡会毁掉整体听感。
- [01:01:37](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3697s) 市面上标称“±3 dB”的传统参数隐藏了宽达 6 dB 的公差陷阱；昂贵的高端或静电音箱在离轴指向与共振表现上可能远逊于设计严谨的平价音箱。
- [01:11:35](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=4295s) 麦克风测出的微小声学干涉（如靠背反射）往往属于人耳和大脑自动容错不计的范畴，盲目依据测量数据做 EQ 纠偏反而会劣化音质。
## Detailed walkthrough
### [00:17](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=17s) 艺术、科学与声音重放的本质
弗洛伊德·图尔（Floyd Toole）开门见山地厘清了艺术、观点与科学之间的界限。艺术与个人观点就像五彩斑斓的光谱，形态各异、随心所欲，没有固定标准；但科学完全不同，科学是非黑即白的客观现实，非存即亡，并且具有可重复性。无论今天、明天，是在本地还是在遥远的廷巴克图（Timbuktu），只要方法正确，科学都能得出一致的结论。尽管科学与艺术截然不同，但二者必须紧密结合，否则人类通过美妙艺术来娱乐大众的追求就会走向失败。舞台上的“声音制作”（sound production）是实时发生的鲜活艺术，任何两场现场演出都不可能完全相同；然而，一旦在前面加上“重”（re-）这个前缀变成“声音重放”（sound reproduction），整个世界便截然不同了——工程与技术的目标是捕捉现场表演，经过艺术化的精心制作与保存，让世界各地的听众在未来的任何时刻按下播放键时，都能听到创作者想要表达的内容。
遗憾的是，这种愿景在现实中极少实现。绝大多数人无论是在家中、车里，还是通过耳机听音乐，所听到的声音都与艺术家最初创作的相去甚远。Toole 将其比作艺术馆的灯光：当人们走进画廊时，展方会确保使用中性的北向自然光（neutral north light），而不是打上刺眼的琥珀色或蓝色滤光，这样才能最真实地还原画作上的本色。声音重放也应遵循这一准则，而不是任由不合格的器材扭曲创作者的意图。
- Floyd Toole 强调：“Most of the time what people hear in their homes, in their cars, through their earbuds or whatever is not what the artists here or any anywhere else created.”（多数时候，人们在家中、车里或通过耳机所听到的声音，根本不是艺术家所创作出来的声音。）
---
### [02:26](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=146s) 声学研究的错配与爱迪生的百年营销戏法
Toole 指出，声学研究领域存在一种荒谬的资源错配：长期以来，大量严谨的研究都集中在大型音乐厅上，还有部分研究针对工作场所（因为这直接涉及人类的职业健康与安全法规）；然而，人们日常耗费最多时间听音乐的场景——家庭、汽车和各类耳机——获得的基础科学研究反而少得可怜。这种研究与现实应用的倒置与大众的娱乐习惯完全脱节，大众虽然会去听音乐会，但绝大部分听音行为实际发生在私人生活空间与交通工具中。
谈到声音重放的历史时，Toole 展示了一张 1918 年托马斯·爱迪生（Thomas Edison）进行留声机宣传的经典照片。当时爱迪生宣称自己发明了一种能够完美模仿人声和小提琴的声音重放设备，并在现场组织了一场所谓的“盲听测试”。Toole 幽默地指出，如果仔细观察照片就会发现，那些受试者头戴的眼罩甚至把一部分耳朵也遮住了，这根本不是科学的盲听，而是一场“既盲又半聋的测试”（blind and partially deaf test）。在此后的一个世纪里，类似的商业演示层出不穷：让真人乐手与扬声器并排演奏，台下观众无不惊呼“根本听不出区别”。但这绝不意味着过去一百年里音频技术没有进步，而仅仅证明了这是一种极其精明、极具煽动性的营销手段（slick marketing / party trick）。特别是在音乐厅等极佳声学环境中，优良的厅堂反射与混响往往掩盖了重放设备的粗糙瑕疵；若换到一个普通房间，这种对比把戏立刻就会露馅。
- Floyd Toole 调侃爱迪生的营销噱头时谈到：“So, it's a blind and partially deaf test. And maybe that's maybe that's how he got away with it.”（所以，这其实是一场盲听兼半失聪的测试。或许他当年就是靠这招蒙混过关的。）
---
### [04:20](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=260s) 从主观臆断到科学测量：消声室与客观规律
真正的听音实验是一项高度严肃、需要严格准备与专注度的艰苦工作，绝不是端着自己喜欢的饮料、坐在沙发上漫不经心地听几首爱听的曲子。如果实验设计严谨、受试条件受控，研究人员就能获得高度可靠的“主观测量”（subjective measurements）。这类主观评价的统计离散度与标准差完全可以达到与客观技术测量相同的精度与一致性。与此相对，仪器的技术测量则必须满足准确与全面两个前提：如果不准确，测量数据就会误导人；如果不全面，就无法反映声音的全貌。这种高门槛的客观测试设施将行业内绝大多数人挡在门外，因为他们根本不具备相应的高精度实验条件。
Toole 回顾了自己的科研生涯，坦言自己非常幸运，无论是在加拿大国家研究委员会（National Research Council, NRC）还是后来在哈曼国际（Harman International），都能使用大型且高精度的消声室（anechoic chamber）进行技术声学测量，并拥有专门建造的受控听音室进行主观评价。当研究者将精准的消声室测量数据与严密的双盲听音评价结果放在一起交叉对比时，客观声学指标与主观听感之间的关联模式便会迅速浮现。这些规律不仅毫不晦涩，甚至“明明白白地站起来对你大声呼喊”：只要在扬声器设计中做对某几项关键的物理特性，听众就会毫不犹豫地给出极佳评价；如果不遵循这些规律，音箱设计就变成了一场完全没有胜算的赌博。
- Floyd Toole 在阐述主客观数据的契合度时指出：“If you do both of those things and start comparing the results, you very, very quickly see patterns, they're not difficult. They're obvious. They just stand up and shout at you.”（如果你同时做好了这两件事并开始对比结果，你会非常迅速地发现其中的规律，它们一点都不复杂，显而易见，甚至直接站起来向你大声呼喊。）
---
### [06:08](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=368s) 盲听实验的方法论：排除偏见与多元对比
音频领域充斥着各种各样的个人观点，不论翻开音响杂志、浏览发烧友网站还是进入网络论坛，各色主观言论铺天盖地。在 Toole 看来，这些脱离科学控制的意见大多只是背景噪音，其中或许有正确的见解，但人们根本无法甄别其真伪。因此，科学的核心任务就是建立严谨的方法，把纯粹的主观意见转化为确凿的客观事实。
要实现这一转变，首要原则就是彻底遮蔽视觉干扰（盲听），遮蔽扬声器的品牌、价格、外形尺寸以及受试者脑海中记诵的器材评论。更进一步，必须实施“双盲”（double-blind）测试：主试人员绝不能通过出题顺序或下意识的举止对受试者产生心理暗示。哈曼的盲听系统完全由计算机程序控制声音序列的随机化，研究人员只需按下启动键便可离开房间去喝咖啡，测试由电脑完全自主运行。此外，所有参与对比的器材必须做极其精准的声级匹配（level matching）。Toole 极其反对传统音频评论员那种“把音箱拿回家听几周”的测试方法，因为这根本学不到任何东西，听觉只能让人逐渐适应器材本身的缺陷；真正的听音辨别只能在即时切换对比中发生。即便做直接对比，简单的 A 对比 B 也有严重漏洞——如果音箱 A 和 B 具有某种相同的频率响应缺陷，听众就很难察觉这一问题。因此必须引入 C 乃至 D 构成多元多路对比（multiple comparison）。Toole 回忆说，早在 1966 或 1967 年他做自己的第一次听音测试时，就很偶然地采用了三路或四路对比的架构，此后几十年这一原则被证明是唯一可靠的方法。
```mermaid
flowchart TD
accTitle: 双盲听音测试核心控制流程
accDescr: 展示双盲听音实验如何通过视觉阻断、声级匹配与四路随机切换来排除主观偏见。
subgraph Controls["物理与声学控制"]
A["声级精确校准 (Equally Loud)"]
B["视觉全遮蔽 (Acoustically Transparent Screen)"]
end
subgraph Logic["计算机随机调度"]
C["完全双盲随机排序 (Double-blind)"]
D["多元即时对比 (A/B/C/D 循环切换)"]
end
subgraph Feedback["数据产出"]
E["受试者独立评分与笔记"]
F["高复现性的主观测量事实 (Facts)"]
end
Controls --> Logic
Logic --> Feedback
```
- Floyd Toole 谈及即时多路对比的必要性时断言：“Only when you compare one thing with another do you start to realize what differences there may be.”（只有当你将一件东西与另一件东西进行对比时，你才会开始意识到它们之间到底可能存在怎样的差异。）
---
### [08:20](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=500s) 单声道评估的优越性与自动化硬件机构
在具体的对比协议中，一个反直觉却至关重要的核心原则是：测试必须以单声道（monophonic，仅使用单只扬声器）形式进行，绝不采用立体声或多声道。多年的心理声学测试表明，单声道测试的灵敏度和揭示力远高于立体声。听众在只聆听一只扬声器时，对音质缺陷的挑剔程度和辨别力达到峰值；一旦开启立体声或多声道，空间声像与通道间的掩蔽效应会迅速分散注意力，声道越多，听众对扬声器本身的缺陷反而越不敏感。此外，在真实的影视与电视重放中，中置扬声器承载了所有对白和绝大部分荧幕画面音效，在整个环绕系统中占据核心地位，其单声道还原能力极为关键。Toole 强调，他们曾多次为了向质疑者自证而追加立体声对比，结果毫无例外：在单声道测试中胜出的扬声器，在立体声测试中必然全盘胜出。在音质保真度这一根本维度上，立体声和多声道没有任何特殊的豁免权。
在具体的测试实现上，每次仅有一位受试者坐在听音室中，手持终端自主切换 A、B、C 或 D，并记录评价。测试音乐被切分成简短片段循环播放，并且每当曲目发生切换，ABCD 字母代号背后对应的实际扬声器就会被电脑重新随机重组，受试者在整场测试中永远无法知晓幕后是哪台设备。在听音幕布的后方，哈曼设计了一套精密复杂的全自动液压驱动扬声器更换系统。三到四只大型扬声器安装在活动导轨上，只需约 3 秒钟，机械装置就能将选中的那只扬声器精准移送至完全相同的物理空间声学中心，排除任何由于摆位角度或离墙距离不同带来的声学偏差。这套机构耗资巨大，但哈曼甘愿为此买单，因为人类的主观心理认知极其脆弱多变，唯有用最高规格的物理与实验工程，才能从多变的人类意见中淬炼出稳固的科学真理。
- 扬声器轮换机构由计算机全自动控制，通过高精度液压机械在 3 秒左右完成声学中心物理替换。
- 循环测试机制：每一段音乐切换时，底层音箱通道重新随机洗牌，等同于开启一次全新的双盲实验。
- Floyd Toole 斩钉截铁地总结单声道测试与立体声的关系：“And every time the the ones that win the mono tests win the stereo tests. There has been no exception to that rule.”（而且每一次，在单声道测试中获胜的扬声器都赢得了立体声测试。这条法则从来没有任何例外。）
### [11:12](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=672s) 哈曼初期的双盲测试挑战与视觉偏见
当弗洛伊德·图尔（Floyd Toole）刚加入哈曼国际（Harman International）时，公司内部甚至没有任何人记得做过哪怕一次盲听测试。当时内部的工程师和技术人员自视为专业人士，他们坚信专业水准足以让自己免受视觉外观的干扰，自认“真正懂得什么是好声音”，因此根本不需要繁琐的盲测筛选。彼时，刚从加拿大国家研究委员会（NRC）和麦吉尔大学（McGill University）毕业不久的肖恩·奥利弗（Sean Olive，现为博士）也加入了哈曼。图尔与奥利弗决定召集这批专业听音员及其他员工进行一系列严格测试。然而过程并非一帆风顺，有几位号称“金耳朵”（golden ears）的核心人物拒绝参与，因为他们的行业声望极高，不愿冒任何可能砸掉名誉的风险。
在正式的测试中，图尔团队选用了四对扬声器：前两对外观完全相同、售价约为 5,000 美元的高端扬声器，唯一的区别仅在于分频器（crossover）设计；第三款是一套售价仅 700 美元的紧凑型卫星加低音炮系统，采用塑料外壳，虽然响度上限不及大箱子，但在正常的前景聆听声压下其声音表现极其出众；第四款则是美国竞争对手推出的一款售价 4,000 美元的扬声器，拥有极具艺术美感的木质雕刻箱体。
在盲听测试中，两款分频设计优秀的扬声器拔得头筹，而小系统与竞争对手的得分则各有高低。然而，一旦撤去遮挡进行明听（sighted test），受试者的评分模式瞬间颠覆：
- **哈曼忠实员工的品牌偏见**：看到自家大型且昂贵的扬声器时，员工们给出了高得多的分数。
- **对廉价小体积的固有偏见**：那套 700 美元的塑料外壳系统即便盲测表现优异，明听时评分却被大幅打低，因为大家从心底里认定“这么小、这么塑料的东西不可能出好声”。
- **外观做工的加分溢价**：竞争对手售价 4,000 美元的木质箱体在明听下得到了视觉加分，得分有所回升。
关于售价 700 美元的小系统，Floyd Toole 评价道：“Disgustingly good for $700. Just disgustingly good.”
```mermaid
flowchart TD
accTitle: 盲听与明听评估偏见对比
accDescr: 展示四款音箱在盲听测试与明听测试下因为视觉偏见产生的评分差异
subgraph Blind["双盲测试 (Blind Test)"]
B1["5000美元音箱 A/B<br/>凭借分频设计客观分化"]
B2["700美元小音箱<br/>声学表现优异获高分"]
B3["4000美元竞品木箱<br/>声音平庸评分垫底"]
end
subgraph Sighted["明听测试 (Sighted Test)"]
S1["品牌与体积加成<br/>哈曼大音箱得分大幅虚高"]
S2["廉价塑料外观偏见<br/>700美元系统评分被骤降"]
S3["豪华木质箱体加成<br/>竞品得分获得视觉补偿"]
end
Blind -->|"移开遮挡幕布"| Sighted
```
---
### [13:55](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=835s) 摆位、录音与听音经验对评估的影响
房间声学对低频响应有着决定性的物理影响，扬声器在房间中的摆放位置以及聆听者的坐姿，都会极大改变人耳接收到的实际声音。在盲听实验中，同一只扬声器只要更换摆放位置，听音员给出的偏好评分就会出现巨大的起伏，四款音箱在不同位置的得分均呈现出明显的离散性。然而不可思议的是，一旦让听音员亲眼看到正在发声的音箱，这种因房间位置引发的听觉差异就奇迹般地消失了。听众的大脑会建立起“既然是同一只音箱，声音必然一样”的心理预期，甚至直接复刻先前的打分，这种认知惯性让听音者仿佛瞬间失聪。
同样的心理效应也表现在音乐素材的选择上。不同录音的频谱平衡各异，乐器编配也不同，因此它们揭示扬声器缺陷的方式也完全不同。在双盲状态下，不同的测试曲目会导致平均偏好评分出现明显波动；而在明听状态下，测试曲目对评分几乎毫无影响，听众抱着“随便播什么我都会打同一个分”的心态完成了评估。针对这种先入为主的认知偏差，Floyd Toole 直言：“If you believe something, you very likely will hear it.” 他指出，发烧友圈子里关于音箱线材等玄学配件的所谓音质提升，完全是视觉与心理暗示构建出的假象，本质上纯粹是商业营销。
经验与人群特征同样深刻影响着主观测试的结果：
- **经验让听评更加挑剔**：长期参与评测、受过专业批判性听音训练的人极难被轻易打动，打分普遍更加严格苛刻；而缺乏经验的听众则倾向于给出更高的整体分数。
- **性别与听力差异**：在相同经验水平下，男性与女性在声音品质的判断上高度一致，击碎了任何性别偏见的假设。唯一存在的统计差异在于，男性由于爱好通常更嘈杂而更容易出现听力受损，这可能导致部分男性听评者给出失真的评分。
---
### [16:59](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1019s) “穿透房间倾听”：人类的听觉场景分析能力
房间声学会彻底扭曲声场，但这是否意味着人耳无法在房间中准确评估扬声器？图尔在离开哈曼前主导设计、并在其离开后完成了一项关键实验：选用三款客观性能高度接近、差异极其细微的顶级扬声器，在完全盲听、随机轮换摆位的严格双盲条件下，分别在三个声学特性截然不同的房间中进行评测。测试结果表明，尽管三个房间的实际声学声场差异极大，但在三个房间中，听音员对这三款扬声器的评分偏好和排序几乎完全一致，分值差异微乎其微。
这一实验揭示了人类大脑一项极其强大的感知机制：人耳具备“穿透房间倾听”（listen through rooms）的能力。在现场音乐会中，观众无需为施坦威（Steinway）或贝森朵夫（Bösendorfer）三角钢琴在不同音乐厅里的表现去逐一“声学均衡”，大脑能够天然地将乐器本身的发声特性与房间反射声分离。面对纷繁复杂的听觉信号，人类大脑必须完成三项极其精密的声学解构：
1. 从复杂的总声场中剥离出扬声器自身的音色特征（timbral signature）。
2. 将扬声器的音色特征与所处房间的声学染色分离开来。
3. 扣除音乐节目源（录音混音）本身的音色风格，而事实上没有任何听音者曾亲临那些商业录音的最终混音现场。
图尔引用了麦吉尔大学校友阿尔伯特·布雷格曼（Albert Bregman）提出的“听觉场景分析”（Auditory Scene Analysis）理论来解释这一现象。在大脑中，我们对发声体（乐器或重放该乐器的扬声器）与环境声学反射进行了分离的“听觉流”（auditory streaming）处理。正如在音乐会上如果低音不足，观众只会归咎于音乐厅声学欠佳，而绝不会责怪低音提琴手或乐器本身；如果一个人在洗手间说话，嗓音因瓷砖混响而变得丰厚，听者依然能完全辨认出那是他的原声并给出极高的保真度认可，同时清楚意识到这仅仅是他在特定空间里的声音。
---
### [21:15](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1275s) 建立主客观关联与扬声器测量的复杂性
在确立了主观测试的可靠性之后，哈曼团队意识到：通过标准化的科学双盲流程，不同人群在不同时间、针对不同节目源所做出的主观扬声器音质评价是完全可复现且高度一致的。基于这一发现，研发团队开始建立一个庞大且不断扩充的主观偏好数据库。
紧接着的关键任务，是建立一套与之并行的技术测量（客观声学参数）数据库。研究人员需要退后一步，将详尽的主观打分与精确的客观物理数据进行横向交叉对比，从而揭示到底是什么样的声学性能指标决定了一款扬声器能够普遍受到大众喜爱，而另一些设计则令人反感。
图尔强调，在开展客观测量之前，必须摒弃一个流传甚广的技术误区：没有任何单一的频响曲线能够完整描述扬声器的声学表现。扬声器是一个全向或半球向的三维声辐射体，它会将声波投射到具有反射面的三维房间中；当这些复杂的直达声与反射声混合并最终到达人耳时，其物理声场结构极其错综复杂，根本不可能单凭在某一固定轴向上测得的一条曲线来还原或评价系统的优劣。
在这段演讲中，弗洛伊德·图尔（Floyd Toole）博士深入探讨了扬声器声学测量的核心逻辑，破除了“在房间聆听位置放置麦克风就能反推扬声器表现”的迷信，并系统介绍了基于消声室测量的标准测试方法及其对实际听音环境的预测能力。
### [22:33](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1353s) 扬声器测量的因果逻辑与 Spinorama 测试方法
图尔指出，许多人试图在房间的聆听位置放一个麦克风，分析采集到的声音并试图反推出扬声器原本发出了什么，但这种逆向推导在声学上是无法实现的。正确的做法是正向推导：在消声室中全面测量扬声器向各个方向辐射的声音，并据此推断声音到达房间内听音位置的状态。
```mermaid
flowchart TD
accTitle: 扬声器声学测试与空间预测流程
accDescr: 展示从消声室Spinorama测量到计算房间曲线及反射声特征的流程
subgraph Chamber["消声室测量 (Spinorama)"]
A["70个空间点位测量<br/>(2米距离/水平与垂直自旋)"] --> B["轴向声 (On-Axis)"]
A --> C["聆听窗口 (±30°水平, ±10°垂直)"]
A --> D["早期反射声 (墙面/天花板/地板反射)"]
A --> E["声功率 (全向辐射能量总和)"]
end
subgraph Metrics["指标计算与评估"]
B & C & D & E --> F["指向性指数 (Directivity Index)"]
D & E --> G["预测房间响应曲线"]
end
subgraph Evaluation["主观体验关联"]
F --> H["音色一致性与平滑度"]
G --> I["房间模式与低频表现 (300-400Hz以上高度吻合)"]
end
```
为了实现这一目标，图尔及团队采用了在消声室内进行的“Spinorama”测试方法。该测试在2米距离下对扬声器进行水平和垂直轨道的旋转，共采集70个测量数据，全面记录其频率响应。这其中包括代表皇帝位直达声的轴向响应（On-Axis），以及覆盖多位听众的“聆听窗口”（Listening Window，水平±30度、垂直±10度）。此外，测试还计算了来自地面、天花板和侧墙的早期反射声，以及无差别向所有方向辐射的能量总和——声功率（Sound Power）。
- **直达声与音色**：最先到达人耳的声音决定了人声或乐器的核心音色特征，因此聆听窗口的响应应尽可能与轴向响应一致。
- **早期反射声**：房间中第二响的声音是墙壁和天花板的早期反射，其音色信息必须与直达声匹配，才能保证整体听感自然。
- **声功率与后期反射**：计算扬声器向所有方向辐射的总体加权能量，用以评估后期反射声场。
> “在所有不同的方向测量离开扬声器的声音，并从这些测量中推测出到达房间内听众位置的声音。这就是我们的做法。” —— 弗洛伊德·图尔
### [25:42](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1542s) 指向性指数与行业标准化的现实阻力
在 Spinorama 数据的基础上，图尔进一步阐述了指向性指数（Directivity Index, DI）的作用。指向性指数为 0 意味着声源呈全向辐射，即向所有方向均匀发声。在低频段，当发声单元的尺寸远小于其辐射声波的波长时，扬声器自然呈现全向辐射特性。例如，20 Hz 的波长约为 57 英尺（约 17.4 米），100 Hz 的波长为 11.3 英尺（约 3.4 米），而低音单元仅约 1 英尺大小，因此即使在 100 Hz 时，低音单元本质上也是无指向性的。
随着频率上升，扬声器的辐射逐渐变得朝向听众并变得狭窄。高品质扬声器的设计目标是在宽频段内保持相对恒定的指向性，只有这样，直达声与反射声的音色才能维持和谐统一。该测量体系已于2013年正式被纳入行业标准，但图尔坦言普及该标准面临巨大的现实挑战：一方面是测试的技术门槛和成本高昂；另一方面是残酷的市场心理——如果测量出的数据不佳，厂商根本不会公开发布。
- **波长与指向性**：低频声波波长巨大，任何小尺寸低音炮都属于全向声源，指向方向对低频辐射没有实质影响。
- **平滑指向性的重要性**：只有扬声器在宽频段内保持均匀平滑的指向性转变，反射声才不会出现频率失真。
- **标准普及困境**：不仅中小厂商缺乏消声室测试能力，更关键的是不良的测量结果往往会被厂商直接隐瞒。
> “我们希望扬声器能够在宽广的频率范围内保持类似恒定的指向性。只有通过这种方式，我们才能实现直达声与反射声音色相匹配的理想表现。” —— 弗洛伊德·图尔
### [27:53](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1673s) KEF 105/2 案例：被忽略的离轴响应与平庸听感
图尔以自己1985年的一项研究为例，展示了一款当时知名的高端扬声器（KEF 105/2）的设计缺陷。该产品的工程团队技术过硬，但他们接到的指令仅仅是“打造尽可能平直光滑的轴向响应”。工程师们成功达成了这一单项目标并举杯庆祝，却彻底忽略了离轴方向的声音表现。
在随后的严谨双盲听音测试中，这款昂贵的扬声器既没有垫底也没有胜出，最终评价仅为平庸的中等水平。图尔通过数据揭示了原因：尽管轴向直达声平直优美，但其在 30 至 60 度离轴方向的表现极为逊色。前四次界面反射（地板、天花板、两侧墙）的能量总和及总声功率均暴露出严重的离轴缺陷，这些有缺陷的反射能量彻底掩盖了轴向平直的优势，赋予了扬声器一种不够中性、不自然的听感个性。
- **设计盲区**：单方面追求完美的轴向响应，而忽视了离轴辐射，往往导致声音在实际房间中表现失真。
- **反射能量的主导地位**：在 10 kHz 以下的宽广频段，人耳接收到的总能量由离轴声音与房间反射共同主导，轴向平直并不能掩盖整体声功率的凹陷或凸起。
- **高低频的不同主导因素**：极低频完全由声功率决定，极高频因单元指向性增强及空气、房间吸音而以轴向声为主，但中间的核心听音区域需要全面考量所有方向的辐射。
> “当我们在双盲听音测试中对比它时，它既没有赢也没有输。它就是一款扎实而平庸的扬声器——而它原本是一款备受推崇的昂贵高端产品。” —— 弗洛伊德·图尔
### [30:54](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1854s) 预测房间响应曲线与房间低频驻波的挑战
图尔展示了基于消声室数据计算出的“预测房间曲线”（Predicted Room Curve）。由于声音随时间推移在空间中不断反射叠加，房间内麦克风所测得的稳态曲线实际上高度受制于扬声器的离轴性能和房间界面的声学传输。
为了验证预测模型的有效性，图尔将扬声器放入实际听音室的左声道位置，对6个座位的听音数据进行平均采样。实验中，扬声器仅仅被移动了半米（18英寸），测得的低频曲线便发生了天翻地覆的变化：底鼓的声音时而浑厚过量，时而几乎消失殆尽。这是典型的小房间驻波（房间模态）导致的恶果。然而，在300至400 Hz以上的频段，实际测量结果与消声室数据推导出的预测曲线高度吻合。这证明只要拥有完整的消声室全方位测量数据，就足以准确预判扬声器在真实房间中中高频的表现。而低频驻波则是所有小房间无法逃避的物理现象，必须采用额外的声学手段进行针对性处理。
| 频段范围 | 主导声学因素 | 消声室数据预测能力 | 听感影响与应对策略 |
|:--- |:--- |:--- |:--- |
| **低频 (< 300-400 Hz)** | 房间边界、驻波模态（Eigenmodes）、声功率 | 无法直接预测房间特定驻波峰谷，但声功率决定低频能量基准 | 决定低频扎实度与底鼓形体感；受音箱与听位摆位影响极大，需专门做低频管理 |
| **中高频 (> 300-400 Hz)** | 扬声器轴向响应、离轴反射声、界面吸收率 | 预测曲线与实际多点平均测量高度吻合 | 决定乐器与人声的音色中性度；取决于扬声器本身的辐射均匀度与指向性控制 |
> “只要高于三四百赫兹，消声室测量数据中所包含的信息就足以准确预测扬声器在房间内的真实表现。” —— 弗洛伊德·图尔
### [33:44](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2024s) 均衡器修不好烂音箱：消声室测量与房间曲线的对应关系
Floyd Toole 强调，完全没有反射的消声室（Anechoic Chamber）听音乐极其难听，房间本身对音乐回放而言是必不可少的好东西。然而，当工程师在一个普通房间里测得一条坑洼不平的频响曲线（幻灯片中的红线）时，如果第一反应是拿起均衡器（Equalizer）把它强行拉平，那就大错特错了。Toole 指出，这么做只会彻底毁掉这只扬声器原本唯一的优点——平坦的轴向响应（On-axis Response）。指向性差、离轴响应崩塌的扬声器，根本不可能靠均衡器来弥补，“你解决不了那个问题；面对那台扬声器，唯一的解决办法就是退货，换一台规矩本分的音箱”。
随即他展示了一只优秀扬声器的测量数据：轴向响应极其平直，离轴响应也极其平滑，指向性指数（Directivity Index, DI）非常优秀。把这只扬声器放进实际房间进行六点平均测量后，会发现房间曲线与消声室中测得的声功率（Sound Power）曲线吻合得不可思议。若把指向性指数上下颠倒（求逆），只要扬声器轴向是平坦的，逆 DI 曲线几乎就等同于声功率曲线。除了在 7 到 8 kHz 以上的高频段高音单元开始产生强指向性、声功率对高音的权重计算略有不足之外，在 200 Hz 以上的大部分关键频段，房间曲线完全可以通过消声室测量数据精准预测出来。
紧接着他又拿出今天在场馆里就摆着的一对现代音箱的实测数据作为反例。这只音箱的轴向响应不平，离轴响应更不平，导致指向性指数充满波动。这种情况下，消声室测出的声功率曲线依旧准确预测了它在房间里的恶劣表现。Toole 借此向听众灌输一个核心观念：消声室里那些看似冰冷抽象的科学图表和数据，绝非脱离现实的数学游戏，它们与人在房间里实际听到的声音有着极为紧密的对应关系。
- 扬声器的两大常见缺陷：一是轴向频响不平（可用 EQ 修复）；二是指向性混乱（EQ 绝对无法修复）。
- 房间内测得的恶劣曲线通常是这两者共同作用的结果，若不知道扬声器本身的消声室参数，单看房间曲线根本无法对症下药。
- 逆指向性指数（Inverse Directivity Index）在扬声器轴向平坦的前提下，是对声功率的绝佳近似估计。
---
### [37:51](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2271s) 自动房间校准的迷思：麦克风不是人脑
提到房间均衡，Toole 询问全场有多少人在家或工作环境中使用房间均衡器，看到只有零星一两个人举手，他调侃道“总得有一个人举手，谢谢你”。他谈到，现在的家用环绕声功放（AVR）普遍内置了自动房间校准程序，或者市面上要卖几百上千美元的独立校准硬件和上门调音服务。这些设备只要按下一个按钮，就会发出怪异的测试脉冲声，随后便向用户拍胸脯保证系统已经完美校准。Toole 直言，这种“通过在未知房间里测量未知扬声器就能给所有人带来完美声音”的宣传完全是虚妄的。
令 Toole 尤为痛心和愤怒的是，诸如 SMPTE、电影工业界、ISO 以及 ITU 等行业权威标准，竟然也在官方规范中推动这种错误的房间测量调音理念。Toole 情绪激动地立下誓言：“在我的风烛残年里，我誓死也要拼尽全力去改变这一现状。”他剖析道，双耳与人脑在房间声场中提取声音特征的机制极为繁复，能够把直达声、早期反射和后期混响拆解开来分别感知，而一只简单的全向测试麦克风加上基础分析仪，只会把所有方向、所有时间的声能无脑叠加。我们在仪器上测出的波形，根本不是耳朵实际听到的声音。
“It can't be true and it isn't true, and in my dying years, I'm going to do my damnedest to change it.” —— Floyd Toole
```mermaid
flowchart TD
accTitle: Room Correction Misconception Flow
accDescr: Diagram showing how single microphone room measurement differs from human binocular hearing perception.
subgraph Ideal["人耳真实听觉感知机制"]
SoundSource["扬声器辐射声场"] --> Direct["直达声 (定位/音色音质主导)"]
SoundSource --> Reflections["房间反射声 (空间感/包围感)"]
Direct --> EarBrain["双耳 + 大脑神经解码"]
Reflections --> EarBrain
EarBrain --> Perception["精准分离感知: 音色优美且真实"]
end
subgraph Flawed["传统全自动室内 EQ 测量谬误"]
SoundSource --> Mic["单一测试麦克风"]
Mic --> Analyzer["频响分析仪 (时间空间无差别叠加)"]
Analyzer --> BlindEQ["粗暴 EQ 逆向补偿"]
BlindEQ --> Ruin["破坏轴向平直响应, 声音恶化"]
end
```
---
### [39:28](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2368s) 过渡频率分界与 Arthur Benade 的启示
Toole 解释了扬声器与房间在不同频段的主导权分工：在低频段，房间驻波和模态（Room Modes）起绝对主导作用；在高频段，扬声器的辐射特性起主导作用。二者之间的交界被称为过渡频率（Transition Frequency / Schroeder Frequency），这个分界点随着房间体积的增大而下移。比如在演讲所在的大厅（Tanna Schulich Hall）中，扬声器在 40 到 50 Hz 以上都拥有绝对的控制权；只有在 50 Hz 以下的极低频段，才需要额外针对房间环境做吸声或低频管理。换言之，在宽敞空间里，只要放进一台真正优秀的高素质扬声器并打开它，除了低频需要稍加注意，听众直接就能获得极高水准的声音体验。
为了印证自己的思考，Toole 提到了已故的著名管乐物理学家兼音乐家 Arthur Benade。Benade 曾在生前给 Toole 寄去了一批自己的学术手稿，其中关于真实演奏与测量的段落让 Toole 醍醐灌顶。Benade 指出，一名乐手在舞台上反复演奏同一个乐句十次，其声学输出是完全一致的，听众席上的人听这十次也觉得毫无二致，哪怕听众在观众席各处走动，听感仍然非常稳定；然而，如果拿着麦克风在观众席不同位置做仪器测量，测出的传输函数曲线彼此之间会有天壤之别，到处充斥着剧烈的梳状滤波（Comb Filter）。
在普通工程师眼里，音乐厅里测出的这种极度混乱的梳状滤波几乎是一场灾难，甚至会得出“这种声场根本无法传递清晰信息、不可能好听”的荒谬推论；但现实是，无数人走进音乐厅聆听现场，奉其为顶级参考体验。这就是“测量到的物理量”与“主观听觉感知”脱节的典型证据——仪器测错了东西。如果换一个思路：只要我们能用消声室测量全面表征声源本身的辐射品质，我们就完全有能力预测听众最终听到的声音表现，而无需过度纠结中间那些因测点微小变动而剧烈跳动的传输函数。
- 物理乐器在房间中被感知为恒定声源，大脑会自动将房间反射声与直达声剥离。
- 现场演奏的逻辑同样适用于高保真回放：优秀的录音 + 优质的扬声器 + 大脑的感知解码 = 还原现场体验。
- 回放链条的成败，一半系于扬声器设计师的技术功底，另一半系于录音行业从业者能否如实捕获并记录乐器与人声辐射的声学特征。
---
### [43:08](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2588s) 科学为艺术服务：音频产业的本质
Toole 话锋转向台下的录音师和音频学生。他承认，录音控制室里摆满了巨大的调音台、繁复的外置机架设备和数不清的旋钮跳线，混音师使用各种声学处理手段去打磨声音，这属于纯粹的艺术创作，完全合法合规。但产业的痛点在于：艺术可以天马行空，工业界却在如何把这份艺术作品原汁原味地交付到消费者手中遇到了巨大的阻碍。
现如今的消费者会在各种场景下听音乐：在家里、在汽车座舱里、通过耳机聆听；录音制品可以实时推流给现场观众，可以做成实体或数字格式随点随播，也可以交付给院线电影系统。Toole 回顾自己一生所处的音频行业，将他的终生哲学总结为一句话：“科学为艺术服务，这就是我们的事业（Science in the service of art is our business）。”
他打趣道，把“科学”、“艺术”与“商业”拼凑在一起其实充满矛盾甚至略带讽刺——科学与商业放在一起常常走偏，艺术与商业在音乐界虽然常见却也冲突不断。但他所供职的硬件制造企业，表面上卖的是耗费木材和钢材、需要花真金白银购买的“大箱子”，但音频行业真正出产的终极产品，其实是跨越媒介、精准传递艺术感动的“好声音”（Good Sound）。
“Science in the service of art is our business.” —— Floyd Toole
### [44:35](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2675s) 困扰音响行业的“混乱之环”与监控困境
声音本身就是最终的产品，正是它决定了消费者脸上是露出微笑还是紧皱眉头。如果声音表现糟糕，消费者绝不会因为音箱外观好看而感到满意。但行业的核心难题在于：我们究竟如何定义“好声音”？通常人们会依靠耳朵去聆听，但一旦单纯依赖主观聆听，就会立即陷入 Floyd Toole 所定义的“混乱之环”（Circle of Confusion）：评价音箱需要依靠录音制品，而这些录音制品却是在录音室的控制室（control room）中制作出来的；在制作过程中，录音师不断挑选、摆放麦克风，摆弄均衡器（EQ），添加时空特效，而他们做出所有这些艺术决策的唯一依据，就是通过控制室里的监听音箱来监听声音。
```mermaid
flowchart LR
accTitle: 混乱之环示意图
accDescr: 展示录音制作与音箱评估之间缺乏客观标准而形成的恶性循环
subgraph Creation["艺术创作端 (录音室)"]
A["录音师使用麦克风/EQ/效果器"] --> B["非标准化的监听音箱"]
end
subgraph Consumption["艺术消费端 (回放)"]
C["高度不一致的商业录音制品"] --> D["家用消费级音箱与房间"]
end
B -->|"决定录音品质"| C
D -->|"用来评价音箱优劣"| A
```
可怕的是，行业内的监听音箱从来没有实现标准化，历年来的品质差异极其悬殊。Floyd Toole 直言不讳地指出，他最近去过的一些控制室里，监听音箱的品质差到如果送给他，他甚至“连给它们腾个放箱子的地方都不愿意”（wouldn't give house room to the speakers）。当艺术家和工程师通过极其糟糕的音箱来创作艺术时，产出的录音制品必然参差不齐。Floyd Toole 举例称，他平时很少用耳机听音乐，但近期为了测试流媒体音乐服务的音量均衡表现而戴上耳机，结果发现虽然各曲目播放响度控制得尚可，但音色平衡（timbral balance）和频谱平衡却差异极大，以至于大多数曲目都需要去调节老式的低音（bass）和高音（treble）旋钮才能入耳。这种不一致的根源就在于监听环境：如果录音室监听音箱自身辐射的低频过多，录音师就会在混音中削减低频，反之亦然。Floyd Toole 强调，创作艺术的领域与欣赏艺术的领域之间必须建立确定的对应关系：
> “专业监听音箱在房间内的表现，必须在某些方面与我们的家用消费级音箱在房间甚至汽车内的表现相似，否则整个系统就无法运转。”—— Floyd Toole
如果缺乏客观测量作为基准，艺术家精心打造的声音就无法准确传达给听众，当听众觉得音乐难听时，根本无法分辨到底是该怪自己的音箱、怪自己的房间声学，还是录音本身就极其差劲。
### [48:02](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2882s) Harman 与《消费者报告》的交锋：从测量到主观打分的数学模型
基于自旋图（Spinorama）数据，声学工程师已经能够在物理层面准确预测音箱在真实房间内的声场表现，但核心挑战在于：这种物理声场能否对应听众的主观感知？这个问题的突破源于 Harman 国际首席执行官打给 Floyd Toole 的一通电话。当时 CEO 质问：“Toole，我们付给你那么多薪水来确保我们的产品听起来出色，那你能不能解释一下最新一期《消费者报告》（Consumer Reports）的音箱评测？为什么我们的一些产品得分那么难看？”Floyd Toole 走进 CEO 办公室坦言，《消费者报告》几十年来在音箱评测上全是在胡搞，行业内深知其结论毫无参考价值。CEO 随即拍板拨款并给予充足时间，要求 Toole “教会他们用正确的方法测试”。
Floyd Toole 与 Sean Olive 随即联手，系统梳理了频谱曲线、客观数据与主观评价之间的心理声学对应关系。Harman 实验室调取了数据库中详尽测试过的 70 款音箱，这些音箱此前均经过严格的双盲、单人独立试听实验，得出 1 到 10 分的主观偏好得分。随后，他们将这 70 款音箱推入消声室进行 Spinorama 测量，并利用算法综合分析轴上与离轴频谱平衡、频宽（bandwidth）、窄带音色缺陷及共振等参数，计算出一个预测分值。计算得出的客观预测分与真实双盲试听得分之间的相关系数高达 0.86。Floyd Toole 调侃道：
> “要是带着这种胜率的几率去拉斯维加斯，赌场用不了多久就会把你轰出门外。”—— Floyd Toole
为了正面回应《消费者报告》，研究团队从这 70 款音箱中单独提取出《消费者报告》刊登评测的原本 13 款音箱。对比结果令人啼笑皆非：Harman 双盲实验中表现最好的音箱，在《消费者报告》的榜单里居然被排在倒数第一，两者甚至呈现微弱的负相关。而当使用 Sean Olive 开发的计算算法处理消声室测量数据时，预测得分与真实盲听得分的相关系数高达 0.995，几乎呈绝对重合的完美线性对应。
- 数据库规模：包含 70 款音箱的消声室 Spinorama 曲线与精密双盲试听打分。
- 全样本预测表现：客观计算预测值与主观听感偏好分的相关系数为 0.86。
- 《消费者报告》13 款比对样本：Harman 双盲试听与 Sean Olive 算法预测得分的相关系数达到 0.995。
### [52:07](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3127s) 电子产品的铁律为何适用于音箱？试听人群的能力差异
针对市场盲从高端评测的现象，Floyd Toole 展示了一个具体的对比案例：4 款售价均在每对 10,000 美元左右的高端音箱，它们在各大发烧友音响杂志上饱受赞誉，任何消费者读完评测都会确信买下它们就等于拥有了行业顶尖水准（state-of-the-art）。但在 Harman 严格的双盲试听中，这 4 款音箱的表现天差地别：仅有两款在争夺第一名，另外两款被远远甩在后面。将它们的 Spinorama 曲线调出对比便一目了然：曲线越平坦、越平滑的音箱，听众打分就越高。Floyd Toole 强调，没有任何人会去设计或购买频响在 10 Hz 到 100 kHz 范围内不平坦的功放或电子元器件，但在音箱领域，很多人却以为可以无视平直频响这一基本法则。双盲试听中的听众用选票证明了物理规律的普适性：平直且平滑的响应能带来更好的听感。
在分析主观测试的可靠性时，Floyd Toole 引用了 Sean Olive 发表在声频工程学会期刊（Journal of the AES）上的研究论文。该研究历时 3 个多月，共有 350 名听众参与，论文涵盖了前 268 名听众的数据，深入研究了听众的经验和专业背景如何影响其评价的一致性（F 统计量，即多次重复盲听同一音箱时打分的稳定程度）和区分能力（能否拉开好音箱与坏音箱的打分差距，而非全给 5 到 7 分的折中分数）。测试结果揭示出音响行业一个极具讽刺意味的现象：
| 听众群体 | 听音辨别与评价表现 | 特征描述 |
|:--- |:--- |:--- |
| **经过选拔与培训的听众**<br>(Trained Listeners) | **最优秀** | 必须首先通过听力筛查（具备正常听觉），随后接受系统训练以精准识别各类声学缺陷；打分一致性最高且区分度极大。 |
| **零售端销售人员**<br>(Retail Salespeople) | **相当出色** | 具备较高的辨别力和评分一致性，表现大幅优于内部营销人员。 |
| **音响媒体测评人**<br>(Audio Reviewers) | **表现欠佳** | 仅有 6 位评论员有勇气自愿参与测试，实际盲听辨别能力并不理想。 |
| **品牌销售与市场营销人员**<br>(Sales & Marketing) | **极其糟糕** | 经常在外宣称自家具备顶级好声音并要求客户买单，但在双盲试听中几乎无法分辨好坏。Floyd Toole 对此戏谑道：“求求你们可别去投票了。”（Please don't.） |
| **普通学生群体**<br>(Students) | **分化明显** | 非音乐学院背景的普通学生缺乏经验，辨别表现各异。 |
### [55:39](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3339s) 人类主观听音评价的一致性与好音箱的标准
Floyd Toole 强调，参与实验的并不是受过专业训练的录音师或音乐系学生，而只是普通的、非音乐非音频背景的大学在校生。然而，尽管这些普通学生在给出打分的稳定一致性以及数值打分的离散跨度上各有差异，但他们最终对音箱产品的优劣排名完全达成了一致。经过长达 40 年的研究，Toole 依然对这一现象惊叹不已：人类在盲听状态下是非常惊人的测量仪器。当身处陌生的房间、听着陌生的音乐、面对未知的扬声器时，只要不让人知道正在听的是什么型号，人们就能给出高度稳定可靠的品质评价；反之，一旦被测者知道正在听的是什么品牌或产品，他们给出的主观意见就毫无价值了。
Toole 回顾了此前提到的理念并指出其完全正确。我们与过去唯一的区别在于，扬声器重放音乐的过程是可以被精确测量的，而乐器本身由于辐射方式五花八门且各具特性，无法以同样的方式建立统一标准。通过将精确的声学测量与心理声学主观评价相结合，团队总结出了制造优秀扬声器的核心准则：轴向频响必须平滑且平直，否则听众一定会挑出毛病；指向性指数（Directivity Index）必须平滑且相对恒定，即使有所变化也必须是渐进、顺滑地过渡。此外，扬声器必须坚决消灭共振，因为人声本质上就是一组共振峰，乐器声也是一堆共振，扬声器一旦带有固有的附加共振，就会对播放的每一段音乐、每一条人声施加同一底色的音染，听众对此极为敏感。
```mermaid
flowchart TD
accTitle: 优秀扬声器的设计准则
accDescr: 展示心理声学与声学测量结合后总结出的优秀音箱关键设计指标。
subgraph CoreRequirements["好扬声器的核心声学要素"]
A["轴向频响 (On-axis Response): 平滑且平直"]
B["指向性指数 (Directivity Index): 平滑且渐变过渡"]
C["完全消除箱体与单元寄生共振 (No Resonances)"]
D["高分辨率测量 (严禁使用 1/3 倍频程分析仪)"]
end
subgraph SubjectiveImpact["听觉主观体验"]
E["避免声音染色 (Coloration)"]
F["宽广且一致的甜点区"]
end
A --> E
B --> F
C --> E
D --> A
D --> C
```
- 为了精准发现并消除共振，必须采用高分辨率测量方式；Toole 调侃道：“如果你手头有一台 1/3 倍频程分析仪，赶紧把它送给你的小孩当玩具吧，因为它根本没有实用价值。”
- Floyd Toole 针对盲听与主观判断的重要性尖锐指出：“And the only way we can get those consistent opinions out of you is to not let you know what you're listening to. Do it blind. But if you know what you're listening to, I don't care what you think.”
---
### [58:46](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3526s) 低频表现的关键权重与传统规格指标的欺骗性
低频表现（Bass performance）在整个主观评价的权重中占据了惊人的 30% 左右。如果低频没有做好，扬声器在其他所有维度的听感表现都会受到严重牵连。低频之所以极难调校，源于人类听觉的等响曲线（Equal-loudness contours）特性：在中频段（约 200–300 Hz 到 3–4 kHz），声压级每增加 10 dB，主观响度才会翻倍；但在低频段（例如 50 Hz），等响曲线急剧密集挤压在一起，可能仅仅 4 到 5 dB 的微小偏差就会让主观响度产生整整一倍的变化。因此，低频响应哪怕只有微小的失真或凸起，都会毁掉整套频谱的平衡。
遗憾的是，消费者在普通零售店里购买音箱时，根本无法从厂家印在包装盒上的传统规格参数中看出这些端倪。绝大多数厂商只会标明“频响范围 58 Hz - 22 kHz，±3 dB”。这整整 6 dB 的公差带极为宽泛（要知道在中频 10 dB 就意味着听感响度翻倍），频响曲线在其中如何剧烈波动、倾斜，消费者一概不知。普通音箱可能只要 200 美元一对，打折促销时甚至 150 美元，买家夹在胳膊底下就抱回家了，但单凭这些传统规格，你根本不知道它的实际声音走向，更无从进行针对性的均衡补偿。
- 低频响应单独占据主观评价加权因素的约 30%，低频调校一旦失衡，整体听感随之崩溃。
- 中频响度翻倍通常需要增加 10 dB，但在 50 Hz 附近的低频区域，由于等响曲线的密集收缩，仅需 4 至 5 dB 就会导致主观响度翻倍。
- 常见的“±3 dB”标称代表着一个宽达 6 dB 的自由浮动区间，隐藏了大量致命的频响起伏与音色倾斜。
---
### [01:01:37](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3697s) 空间平均测量（Spinorama）与不同价位音箱的真实表现对比
一旦使用完整的全套测量数据（Spinorama，自转测量），音箱的所有声学缺陷都会暴露无遗。如果在各个空间平均曲线上都出现了同样的小隆起，那就证明存在共振。扬声器换能器属于最小相位器件（Minimum phase devices），这意味着只要振幅频响曲线完全平直且没有任何突起，就不会存在共振发响；反之，一旦频响有突起，就必然伴随着瞬态共振（Ringing）。如果内置均衡器能够精确抹平这些突起，共振拖尾就会被一并消除。这正是内置放大器和数字信号处理的现代有源监听音箱（如 JBL M2）的巨大优势所在，这也是音箱产业制造真正顶级产品的必经之路。
```mermaid
flowchart LR
accTitle: 最小相位特性与共振抑制逻辑
accDescr: 展示扬声器作为最小相位系统，频响凸起与瞬态共振消除的物理关联。
subgraph AcousticIssue["物理缺陷"]
R1["频响曲线上出现凸起 (Bump)"] --> R2["系统产生共振与鸣响 (Resonance & Ringing)"]
end
subgraph Correction["DSP 与有源校准"]
C1["精确匹配滤波 EQ"] --> C2["抹平频响凸起"]
end
R1 -.->|"通过测量定位"| C1
C2 --> C3["共振与拖尾彻底消除"]
```
接着，Toole 对比了市场上几种不同价位音箱的实测 Spinorama 数据，揭示了价格与声学品质之间的巨大脱节：
| 音箱类型 / 价格 | 标称规格 | Spinorama 实测表现与缺陷 | 听觉后果与评价 |
|:--- |:--- |:--- |:--- |
| **入门廉价音箱**<br>（约 200 美元/对） | 58 Hz - 22 kHz，±3 dB | 频响整体倾斜，存在多个微小共振突起；曲线散漫占用 6 dB 公差空间。 | 表现平庸，但符合该价位预期；若知道具体频响曲线，尚可通过音调控制手动倾斜拉平。 |
| **高素质高性价比音箱**<br>（约 1,800 美元/对） | ±3 dB（市场部保守标注） | 实际表现远优于标称；频响极其平滑平直，几无寄生共振。 | 达到边际效益递减点，声音表现无限逼近业界极限；比售价高其 10 倍的产品更为准确。 |
| **静电混合落地音箱**<br>（约 11,000 美元/对） | ±3 dB | 大振膜静电面板叠加动圈低音；轴向与听音窗口曲线分离严重，充斥密集的寄生共振。 | 极具工业美感却声学极其糟糕；指向性极窄，沙发上并排坐两个人听到的声音都完全不同。 |
| **超昂贵发烧音箱**<br>（约 24,000 美元/对） | ±3 dB | 用满了 6 dB 容差；不仅存在清晰可闻的强共振，而且在指向性上有严重突变缺陷（忽宽忽窄）。 | 无法通过常规均衡手段纠正，中高频音色杂乱，在声学精度上完败给 1,800 美元的产品。 |
| **高精度监听音箱**<br>（约 16,000 美元/对） | ±1 dB | 频响控制极其严苛，在极窄的公差范围内艰难达标。 | 面临极具挑战性的技术极限，做工和声学公差控制极其严苛。 |
针对市场上对高价静电音箱盲目推崇的现象，Toole 进行了无情驳斥。许多发烧友沉迷于“静电薄膜几乎没有质量，因此启动奇快、无比精准”的营销话术，但全角度实测表明，这类大面板静电单元指向性过窄，在多个频率上产生剧烈鸣响，甚至让同一张沙发上的两个人听到完全不一样的声音。而对于标价 24,000 美元的系统，其实测居然存在一个明显高于听阈的共振峰，指向性也存在严重凹折，根本无法利用 EQ 修复。相比之下，一对优秀的 1,800 美元音箱在音色准确度上足以彻底碾压这些标价高出十倍的昂贵产品。
- Floyd Toole 针对 1,800 美元优质音箱的极致性价比给出了极高的评价：“So if you were to pick ask me what the point of diminishing returns would be that well clearly that's an example of it because honest to God you could pay 10 times that or more and not get as much in terms of timbral excellence and accuracy as that portrays.”
- 针对 11,000 美元静电面板音箱的营销神话，Toole 毫不客气地拆穿其虚伪面具：“Well, no, it's not. It's not fast. It's not accurate. And it rings like hell at a whole bunch of frequencies because it's very directional.”
### [01:06:05](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3965s) 音箱行业生产一致性的挑战
Floyd Toole 强调，在大规模生产中保持音箱的一致性是一项艰巨的任务。他认为，“One of the fundamental problems of the loudspeaker industry is that consistency is one thing to design a golden prototype and you can you know you might get lucky and do really good job on that. It's a very different thing to clone that in mass production.” 他指出，消费者在购买时往往并不清楚自己买到的产品是否与原型一致，尤其是在面对那些缺乏质量控制的小型公司时。对于 Harman 等公司而言，花费高昂成本在生产流程中确保产品符合规格，正是消费者所支付价值的一部分。
Toole 还风趣地批评了行业现状，称关于音箱规格的说明书甚至不如轮胎侧面的信息有用，这种“信任我（Trust me）”式的营销充斥在消费级和专业级设备领域，缺乏实际且有意义的技术指标。
### [01:07:52](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=4072s) 克服“混乱循环”与中性音箱的重要性
在讨论为何好音箱并不总能保证好声音时，Toole 提到了“混乱循环”（Circle of Confusion）。他解释说，由于录音环节的极高变量（不同的录音室、不同的监听设备、不同的混音环境），即使拥有完美的重放音箱，也无法弥补源头的差异。他提出，解决之道在于生产环节和家庭重放环节都使用“中性”（neutral）音箱。
他以“清晰的窗户”作为比喻，指出艺术家的工作是创造窗外的风景，而音箱应该是那扇清晰的窗户。如果窗户本身是有色的（非中性），那么每个人看到的景象都会不同，这会导致对录音质量的误判。他乐观地表示，随着测量技术的普及，中性音箱的价格正在降低，即使是 200 美元的入门产品配合低音炮，也能达到相当不错的水准。
### [01:11:35](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=4295s) 最终案例分析：测量数据与真实听感的偏差
在结束演讲前的最后一个演示中，Toole 展示了 JBL M2 音箱在不同环境下的测量曲线，包括家庭影院及容量从 24 到 516 座位的多个电影院。他指出，在 200Hz 以上，这些环境的测量表现高度一致。
他特别解释了测量中的一个小插曲：部分曲线在测量中出现异常波动，是因为麦克风放置在靠近座椅靠背的位置，引发了声学干扰。他强调，“It's an acoustical interference thing that the human two ears and a brain just they don't we don't hear it.” 这种现象如果不加甄别，容易诱导技术人员对其进行不必要的均衡调节（Equalization），从而“Making what might have been a really good loudspeaker sound worse”（让原本优秀的音箱听起来更糟）。
- **关键观点**：在 200Hz 以上的频率，通过测量手段能够有效达成重放的一致性，但必须小心剔除由于测量位置引起的“听觉无法感知”的声学异常。
- **结论**：人类的听觉系统（大脑与双耳）具有一定的容错机制，盲目根据测量数据进行修正往往会产生反效果。
## Notable quotes
> [13:55](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=835s) If you believe something, you very likely will hear it. — Floyd Toole
> [55:39](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=3339s) And the only way we can get those consistent opinions out of you is to not let you know what you're listening to. Do it blind. But if you know what you're listening to, I don't care what you think. — Floyd Toole
> [08:20](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=500s) And every time the the ones that win the mono tests win the stereo tests. There has been no exception to that rule. — Floyd Toole
> [06:08](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=368s) Only when you compare one thing with another do you start to realize what differences there may be. — Floyd Toole
> [43:08](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2588s) Science in the service of art is our business. — Floyd Toole
> [30:54](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1854s) 只要高于三四百赫兹，消声室测量数据中所包含的信息就足以准确预测扬声器在房间内的真实表现。 — 弗洛伊德·图尔
> [25:42](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=1542s) 我们希望扬声器能够在宽广的频率范围内保持类似恒定的指向性。只有通过这种方式，我们才能实现直达声与反射声音色相匹配的理想表现。 — 弗洛伊德·图尔
> [44:35](https://www.youtube.com/watch?v=zrpUDuUtxPM&t=2675s) 专业监听音箱在房间内的表现，必须在某些方面与我们的家用消费级音箱在房间甚至汽车内的表现相似，否则整个系统就无法运转。 — Floyd Toole
## Mentioned
Floyd Toole, Sean Olive, Arthur Benade, Albert Bregman, Thomas Edison, Harman International, National Research Council (NRC), McGill University, CIRMMT, Consumer Reports, KEF 105/2, JBL M2, SMPTE, ISO, ITU, Audio Engineering Society (AES)
## Video description
To view a visualization of the videos recorded by CIRMMT, visit https://www.cirmmt.org/en/videos
CIRMMT Distinguished Lectures in the Science and Technology of Music
Floyd Toole, consultant to Harman International, USA
16 April 2015 - Tanna Schulich Hall
http://www.cirmmt.org/activities/distinguished-lectures
https://www.facebook.com/CIRMMT/
APA video citation:
Toole, F. (2015, May 14). Sound reproduction - art and science/opinions and facts -
CIRMMT Distinguished Lectures in the Science and Technology of Music. [Video file].
Retrieved from https://www.youtube.com/watch?v=zrpUDuUtxPM
[Source](https://www.youtube.com/watch?v=zrpUDuUtxPM)
Source quotes
[13:55] Floyd Toole: If you believe something, you very likely will hear it.
[55:39] Floyd Toole: And the only way we can get those consistent opinions out of you is to not let you know what you're listening to. Do it blind. But if you know what you're listening to, I don't care what you think.
[8:20] Floyd Toole: And every time the the ones that win the mono tests win the stereo tests. There has been no exception to that rule.
[6:08] Floyd Toole: Only when you compare one thing with another do you start to realize what differences there may be.
[43:08] Floyd Toole: Science in the service of art is our business.
[30:54] 弗洛伊德·图尔: 只要高于三四百赫兹，消声室测量数据中所包含的信息就足以准确预测扬声器在房间内的真实表现。
[25:42] 弗洛伊德·图尔: 我们希望扬声器能够在宽广的频率范围内保持类似恒定的指向性。只有通过这种方式，我们才能实现直达声与反射声音色相匹配的理想表现。
[44:35] Floyd Toole: 专业监听音箱在房间内的表现，必须在某些方面与我们的家用消费级音箱在房间甚至汽车内的表现相似，否则整个系统就无法运转。
Source quotes
[15:26] Floyd Toole: If you believe something, you very likely will hear it.
[66:05] Floyd Toole: there's more useful and reliable information on the side of a tire than there is in a loudspeaker specification sheet.
[55:39] Floyd Toole: The old idea that third-octave data is useful—I'm afraid if you have a third-octave analyzer, give it to your kids because it's a toy.
[20:11] Floyd Toole: We have to admit right away that there is no single curve, no frequency response curve or whatever curve you choose to identify it as, that can completely describe the performance of a loudspeaker.
[42:18] Floyd Toole: Science in the service of art is our business.
Original transcript
[0:17] Good afternoon everyone and uh what can
[0:20] I say after that?
[0:22] Uh I'm going to try to hush along be a
[0:26] little hasty here because uh times have
[0:29] passing and we have as I say some sound
[0:32] production coming after us here.
[0:36] Art and opinions are shown here in
[0:38] multiple colors and shades because
[0:40] that's really what they are. It's
[0:42] whatever you want it to be. Whatever art
[0:44] is whatever you want it to be. Opinions
[0:46] are whatever they happen to be. But I
[0:49] come from the science side of things.
[0:51] It's black and white. It's reality. It's
[0:53] there or it's not there. It's repeat. If
[0:55] it's good, if it's accurate, it's
[0:56] repeatable. You can do it today,
[0:58] tomorrow, here, or in Timbuktu and get
[1:01] the same answers. Science is different
[1:03] from art. Science is different from
[1:05] opinions. And yet they go together. They
[1:09] must go together. In fact, if they
[1:11] don't, we we will fail at our our
[1:15] enterprise of entertaining the world
[1:16] with with wonderful art, sound
[1:19] production, the real thing happening on
[1:22] stage like this. It's it's it is art in
[1:25] real time. And it'll do two performances
[1:27] may never be exactly the same. That's
[1:29] just what re That's what it is. Sound
[1:32] reproduction. You put two letters in
[1:33] front of it. re and the world changes
[1:36] because you're going to try to capture a
[1:39] performance
[1:40] artistically craft it and save it so
[1:45] that somebody somewhere sometime can
[1:47] press a play button and if everything
[1:50] works right they will hear what you've
[1:52] created. Sadly that is rarely the case.
[1:56] Most of the time what people hear in
[1:58] their homes, in their cars, through
[2:00] their earbuds or whatever is not what
[2:03] the artists here or any anywhere else
[2:06] created. I think it's a shame. I think
[2:10] it's nice when you know you go to an art
[2:12] gallery, they make sure that the
[2:13] lighting is correct so that the colors
[2:15] on the paintings will be what as close
[2:17] as it possible at least what the artist
[2:19] created. They don't put amber lights on
[2:22] or blue lights on. They put neutral
[2:24] north light on.
[2:26] What's wrong with this picture? Well, it
[2:28] just simply says that a lot of research
[2:31] has been done on concert halls, some on
[2:34] work spaces because human health is
[2:36] involved. Homes and cars and headphones,
[2:39] not so much. But where do we listen?
[2:42] It's the other way around. That's where
[2:44] we are entertained. We go to concerts,
[2:47] but not that often.
[2:49] Uh we listen at home. We listen in our
[2:51] cars. We listen through headphones. I
[2:52] don't know how much but certainly the
[2:55] trend is right
[2:57] and in fact if you if you think about
[2:58] marketing of of products and sound I
[3:01] mean if you think of you look at that I
[3:04] love that because here is 1918 is Thomas
[3:07] Edison claiming that he has created a
[3:10] reproduction device that will imitate a
[3:13] human being or a violin and this is a
[3:17] blind test
[3:19] but you'll notice that some of the
[3:20] blindfolds cover the ears
[3:24] So, it's a blind and partially deaf
[3:27] test. And maybe that's maybe that's how
[3:29] he got away with it. But no, back back
[3:32] then and between then and now, there
[3:34] have been several demonstrations of per
[3:38] live performers beside loudspeakers,
[3:41] reproducing live performers, and the
[3:44] audiences would be amazed that they
[3:46] couldn't tell the difference. Well, you
[3:48] wonder, you know, have we made no
[3:50] progress in the hundred years since
[3:53] since Thomas Edison? Of course, we have.
[3:56] Uh, it's just that it's slick marketing.
[3:59] It's a good party trick. Let's just say
[4:01] that. Uh, in the case of the concert
[4:03] hall demonstrations, I might just add, I
[4:06] suspect there's good strong evidence to
[4:08] indicate that it's the hall that makes
[4:10] the difference. If they were to do that
[4:13] test in anything other than a really
[4:15] good hall, it wouldn't perform. wouldn't
[4:17] resolve wouldn't work as well. Okay, if
[4:20] you got science involved
[4:22] and we have to pay pay serious
[4:24] attention. We're going to do listening
[4:25] tests. They're not just sitting around
[4:27] with your drink of choice with listening
[4:30] to your music of choice. It's actually
[4:32] hard work. You have requires preparation
[4:35] and concentration.
[4:37] And if you do it right under the right
[4:39] circumstances, you get something that I
[4:41] call subjective measurements because the
[4:43] results from the listeners are highly
[4:46] repeatable, very consistent with
[4:48] standard deviations of the same kind as
[4:51] you get out of technical measurements.
[4:54] If you're going to do technical
[4:55] measurements, they have to be accurate,
[4:56] otherwise they lie to you. And they have
[4:59] to be comprehensive, otherwise they're
[5:00] not telling you all of the truth. And so
[5:03] uh that is a a requirement that
[5:06] eliminates an awful lot of people in our
[5:08] business from the game because they
[5:10] can't they don't have the facilities to
[5:12] do these these highly precise technical
[5:15] measurements. I have been fortunate both
[5:18] at the National Research Council and at
[5:19] Harmon to have had access to wonderful
[5:23] large accurate anacic chambers to make
[5:25] measurements and dedicated listening
[5:27] rooms to do subjective evaluations. And
[5:30] if you do both of those things and start
[5:32] comparing the results, you very, very
[5:35] quickly see patterns, they're not
[5:37] difficult. They're obvious. They just
[5:39] stand up and shout at you. That there
[5:41] are certain things you do in a
[5:42] loudspeaker. And if you do those, people
[5:44] give it the thumbs up right away. They
[5:46] love it. If you don't do those things,
[5:49] then you're just taking the big gamble
[5:51] you don't want to take. So, let's talk
[5:54] about listening tests first. It's the
[5:56] single most important and really the
[5:58] most difficult factor. uh in in
[6:01] achieving this understanding of the
[6:03] relationship between what we measure and
[6:05] what we hear. There are lots of opinions
[6:08] around. My goodness, just pick up any
[6:10] audio magazine, go on to any audio
[6:12] website, enter any forum on the topic,
[6:15] you'll find opinions of all colors and
[6:18] flavors.
[6:20] And uh they really just are noise in the
[6:23] background. And some of them may be
[6:24] right, but you don't know which ones. So
[6:27] we need methods that turn opinions into
[6:29] facts.
[6:31] First thing we do is we make it blind
[6:33] obvious. You just disguise what the
[6:35] person is listening to so they can't see
[6:37] the brand, the price, the size, and
[6:40] remember what their favorite reviewer
[6:42] said about it, what have you. That kind
[6:43] of thing. Make the test double blind
[6:47] if you're a scientist. So then you know
[6:49] that you the experimentter, me the
[6:51] experimentter, I am not influencing the
[6:54] result by the sequence in which I'm
[6:56] presenting these sounds or anything as
[6:59] obvious as that. That has to be
[7:00] randomized and in fact the tests that
[7:02] we've done for years are randomized by a
[7:05] computer. We just press go and go have a
[7:07] cup of coffee. Test is done by the
[7:09] computer. So it is truly double blind.
[7:12] You have to make sure the the test sorry
[7:14] the the sounds you're comparing are
[7:16] equally loud. where you do a multiple
[7:19] par multiple comparison test so you're
[7:21] not just comparing one with another.
[7:23] It's certainly not the take-home and
[7:24] listening listen to it kind of test that
[7:26] reviewers do because that teaches
[7:28] nothing. You you learn really nothing
[7:30] from that except that you might be able
[7:34] to adapt to its problems or not adapt to
[7:37] its problems. Only when you compare one
[7:39] thing with another do you start to
[7:41] realize what differences there may be.
[7:45] And if you do a simple A versus B
[7:47] comparison and both A and B share the
[7:50] same problem, you won't hear that
[7:52] problem. So, it's nice to have C and
[7:54] maybe D to to mix things up a little
[7:57] bit. It uh it seems obvious looking back
[8:02] on it. All I can say is that I'm very
[8:05] fortunate that the first listening test
[8:06] I ever did in about 1966 or ' 67 was a
[8:11] three or four-way test and I've just
[8:12] been doing it ever since. But now I we
[8:16] we know that it's probably the the
[8:18] preferred way of doing things. So the
[8:20] listener controls which one of these
[8:22] speakers is active. And uh um when
[8:25] you're doing these comparisons, you do
[8:27] them one at a time in monophonic one one
[8:31] speaker, not stereo, not multi-
[8:33] channelannel. Uh why? Well, because it
[8:36] just turns out that monophonic
[8:38] comparisons are much more revealing.
[8:40] People are much much fussier about what
[8:44] they can hear and not and like and don't
[8:47] like when they're listening just to a
[8:49] single loudspeaker than they are when
[8:51] they're listening to stereo or multiple
[8:53] channels. Multiple channels. In fact,
[8:54] the more channels are active, the less
[8:56] critical the listener is of the
[8:58] loudspeaker itself. But we have to pay
[9:01] attention to mono because it's a reality
[9:04] in television and movies. Most of the
[9:06] sound comes monoponically from the
[9:08] center front loudspeaker.
[9:10] It is the most important bout speaker in
[9:13] the whole room because all the dialogue
[9:16] comes from it. Most of the onscreen
[9:19] sounds come from it. Only in high budget
[9:22] blockbuster movies do they bother to pan
[9:24] anything. So we do the test in mono
[9:28] and we occasionally have done
[9:31] comparisons in stereo as well just to
[9:33] prove the point to uh questioning
[9:36] persons. And every time the the ones
[9:40] that win the mono tests win the stereo
[9:42] tests. There has been no exception to
[9:44] that rule. There is nothing special
[9:46] about stereophony or multi- channelannel
[9:49] so far as sound quality is concerned.
[9:53] So here's a person one person at a time.
[9:57] They just sit there and listen to music
[9:58] and press which of A, B, C, or D they
[10:01] want to listen to. They make notes. The
[10:04] notes are are compiled in the computers.
[10:08] The music is in loops. So it's it's easy
[10:11] to make these comparisons over a period
[10:13] of time. Every time the music changes,
[10:17] the speakers associated with each of
[10:19] with each of the ABCD codes also
[10:21] changes. So each time the music changes,
[10:23] it's a whole new experiment. Person see
[10:26] at no time knows what's behind the
[10:27] screen. Here I've removed the screen in
[10:30] the bottom picture and you can see that
[10:33] uh there are three loudspeakers in that
[10:35] test. Only one of them is in the forward
[10:38] active position. It takes about 3
[10:41] seconds to exchange. It's all
[10:43] hydraulically driven and computer
[10:45] controlled. Cost a fortune.
[10:49] Uh our harman paid willingly. It's like
[10:52] why bother? Why do we bother spending
[10:54] that money going through all this
[10:56] rigmmoral of blind and double blind and
[10:59] all that sort of thing? Well, because it
[11:01] matters. People are are fickle.
[11:05] Humans are desperately fickle. We we
[11:08] offer up opinions of what we're
[11:09] listening to which are different if we
[11:12] see what we're listening to. And to
[11:15] prove that point, when I joined Harmon,
[11:17] nobody there could ever remember having
[11:19] done a blind test. Not even d not a
[11:21] single blind test even they were all
[11:24] professionals I was told you know and as
[11:26] a professional they're not biased by
[11:29] what they see they they really know what
[11:31] good sound is and so because they know
[11:34] what good sound is they they can do
[11:36] these tests without any complications of
[11:39] blind screening so on so uh Shawn had
[11:44] just joined the company at the time Sean
[11:46] Olive my colleague from NRC and McGill
[11:50] U
[11:51] now Dr. Shaol anyway uh what he uh what
[11:55] we did was just to bring in some of
[11:58] these professional listeners and some
[12:02] others and put them through a battery of
[12:05] tests. Uh not everybody wanted to do it.
[12:09] There were a couple of holdouts. They
[12:10] just did they felt that they had nothing
[12:12] to lose if their reputa you know they
[12:14] their reputations as golden ears were so
[12:17] high that uh they didn't want to
[12:20] threaten that reputation. Anyway, in
[12:22] these blind tests we had four
[12:23] loudspeakers. Two were rated very
[12:25] highly. These were visually identical
[12:27] speakers. They differed only in their
[12:29] crossover designs. Uh this uh where's
[12:34] the arrow? There it is. This test, the
[12:36] speaker down here, the little is a
[12:38] little $700 subset system. Disgustingly
[12:41] good for $700. Just disgustingly good.
[12:44] It wouldn't play as loud as the other
[12:45] ones, but if you kept the sound levels
[12:48] at, you know, good foreground listening
[12:50] level and not pushing it too hard, it it
[12:53] just it cooked. It was a good system for
[12:55] 700 bucks. These others are 5,000. So,
[12:58] here's a US competitor. beautiful wooden
[13:01] cabinet, sculptured looking, wonderful,
[13:04] $4,000.
[13:06] And in the blind tests, we had two
[13:08] winners and two not so good ones. We let
[13:11] them see what they're listening to. And
[13:13] the loyal Harmon employees gave these
[13:15] ones even higher scores, of course,
[13:17] because they were beautiful and they
[13:18] were big and they were Harmon products.
[13:21] Uh, this was this little one was a
[13:23] Harman product, too. But they could
[13:25] never overcome the the the bias that it
[13:29] was so small and plastic that it
[13:31] something small and plastic could not
[13:33] sound good and so they reduced its
[13:35] score. The competitor did a little
[13:38] better because they could see it now and
[13:40] and it had this beautiful cabinet and so
[13:42] they gave it a little more but even then
[13:44] they they felt that it uh wasn't quite
[13:46] up to the the Harmon product. So you
[13:50] know we get different opinions when you
[13:51] see the product when you don't see the
[13:53] product.
[13:55] Another fact is that the room matters
[13:58] especially at low frequencies it matters
[14:00] a huge amount. Where you put the
[14:03] loudspeaker in the room and where you
[14:04] sit in the room changes what you hear.
[14:08] And so if you move the speakers around
[14:10] and do the test we should get different
[14:12] opinions when the speakers are in
[14:13] different places. Uh here on the left
[14:16] you see with the black speaker
[14:18] in the uh blind tests in two locations
[14:21] you get two very different ratings. Huge
[14:24] difference. The white speaker two two
[14:26] same speaker two different locations
[14:29] different ratings. This one same speaker
[14:31] different locations different ratings.
[14:33] Same speaker different locations
[14:34] different ratings. You let them see the
[14:36] speakers and suddenly those differences
[14:39] disappear.
[14:41] They say well it's the same speaker.
[14:42] It's got to be has to sound the same.
[14:44] And I gave it a, you know, 5.6 the first
[14:46] time. I'll give it a 5.6 the second
[14:48] time. They they just kind of go deaf.
[14:53] Uh different recordings matter. I mean
[14:55] recordings are not spectally balanced in
[14:58] the same way. They don't have different
[15:00] the same instrumental content and so
[15:01] they will reveal problems in
[15:03] loudspeakers in different ways. And
[15:05] that's why we use a bunch of different
[15:07] recordings. And here we have in the
[15:09] blind tests the black lines showing that
[15:12] the choice of recording affected the the
[15:15] average preference rating. But in the
[15:17] cited test didn't make a difference.
[15:20] Play me anything. I'll give you the same
[15:22] number.
[15:24] It's interesting.
[15:26] And experience makes listeners fussier.
[15:29] And this has gone on for years. because
[15:31] we noticed this a long time ago that the
[15:33] people who are involved with the program
[15:36] and and do a lot of critical listening,
[15:40] uh they're not as easily impressed as
[15:42] people without that kind of background
[15:44] experience. And so the scores from the
[15:46] inexperienced listeners go up and you'll
[15:48] see that the males and the females
[15:50] agree.
[15:52] So there anybody who has entertains
[15:55] notions of sexism, you can forget them.
[15:57] And this is not the first time I've done
[15:59] this kind of test. the the listeners of
[16:02] both sexes are just fine except that
[16:03] more males suffer from hearing loss than
[16:06] females and so there's the probability
[16:08] that you'll have some faulty opinions
[16:10] from people from males with hearing loss
[16:13] we just enjoyed louder hobbies I guess
[16:17] you know so when you do cited
[16:19] evaluations when you know what you're
[16:21] listening to even if you can't lay eyes
[16:23] on it if you just know what you're
[16:25] listening to uh your bias your opinions
[16:29] are biased even before the first note of
[16:31] music has been played. If you believe
[16:33] something, you very likely will hear it.
[16:36] And this happens, you know, those
[16:38] anybody here is an audio file or reads
[16:40] the audio file press will know
[16:42] immediately what I'm saying. I mean,
[16:44] this is why these are the imag imagined,
[16:47] not real imagined differences between
[16:50] speaker wires and that kind of thing.
[16:52] They're not real. They they are it's
[16:54] complete marketing.
[16:59] Is the room a fundamental problem? Well,
[17:01] uh I one of the last things that was
[17:04] done before I left was to prepare this
[17:07] experiment and it was done after I left
[17:10] uh where we compared three really good
[17:13] loudspeakers. They were closely rated
[17:15] there. The differences were subtle. You
[17:18] had to pay attention
[17:20] to make the test a difficult test. And
[17:22] so we compared these three loudspeakers
[17:24] in three different rooms completely
[17:26] blind. There was a screen they listeners
[17:28] couldn't see what they were hearing. And
[17:31] we randomized the positions and did it
[17:33] several times. It was all done properly
[17:35] balanced from a scientific point of
[17:37] view. And the upshot of it was that in
[17:40] each of these three very different
[17:42] rooms, the listeners were able to rate
[17:44] the loudspeakers and they rated them in
[17:46] almost precisely the same way, certainly
[17:48] the same order.
[17:50] uh the numbers changed a little bit but
[17:52] only a little bit. It's it's remarkable
[17:54] how small the differences were when you
[17:58] think of how large the acoustical diff
[18:01] soundfields
[18:03] differ from one another. How greatly
[18:05] they differ from one another.
[18:10] So we conclude that we have some ability
[18:12] to listen through rooms
[18:15] happens in live performances. you we
[18:18] don't have to equalize a strat a
[18:21] Steinway or Bosendorf or piano in for
[18:25] for whichever the room it's in. It's
[18:27] it's the speaker the loudspeaker or it's
[18:29] the piano or whatever it is in the room
[18:34] and we can separate the sound of the
[18:37] piano or the voice or whatever it is
[18:39] from the room to a great extent.
[18:43] And when you think about what's going on
[18:46] that what has to be going on in this
[18:48] brain between our two ears,
[18:51] it it's it's a remarkable thing because
[18:55] based on
[18:57] masses of 40 years of doing this, I can
[19:00] say that in order to give me or us any
[19:04] experimentter consistent reliable
[19:07] opinions about the sounds that these
[19:10] people are hearing, they will have had
[19:12] to do the These things they will have
[19:14] had to extract from the sound field the
[19:18] tambberal contribution from the
[19:19] loudspeaker so that they can comment on
[19:21] the loudspeaker.
[19:23] And to do that they have to separate the
[19:26] loudspeaker tambber tambberal signature
[19:28] from the room.
[19:30] And in order to make sense of either of
[19:33] those they will have had to take into
[19:35] account the tambal signature of the
[19:37] program the recording.
[19:40] And believe me, nobody has ever would
[19:43] could ever claim to have been at the
[19:46] final mix of the recording they're
[19:47] listening to.
[19:49] So, how on earth is this possible? How
[19:52] can people, even inexperienced people do
[19:57] this? How can experienced people do
[19:59] this? This is this is one heck of a heck
[20:02] of a party turk really and truly. Uh and
[20:06] in the end after you know p pondering
[20:09] this for years
[20:11] it has to be you know uh Bregman another
[20:15] McGill alumni
[20:18] uh his auditory scene analysis. It has
[20:20] to be that we are streaming in our minds
[20:24] we are separately streaming the sound of
[20:27] the musical instrument or the
[20:28] loudspeaker reproducing that musical
[20:30] instrument from the room. They're
[20:33] separate. If you go to a concert and the
[20:36] bass is weak, you don't blame the bass
[20:39] instrument or the musician playing the
[20:41] bass instrument. You blame the concert
[20:43] hall.
[20:45] So, we are able to do that and we do it
[20:47] just subconsciously.
[20:49] If I my voice has a sudden tambal
[20:53] richness, you may recognize that we're
[20:55] in the men's room. And you know, it's
[20:57] because of the reverberation in the
[20:58] room. It's just it's still me. On a
[21:00] scale of 10, the fidelity rating for my
[21:03] voice is is going to be 10. But it's
[21:06] just Floyd in this room as opposed to
[21:08] Floyd in a different room. So we we seem
[21:11] to do this. And this is an amazing
[21:13] amazing feat.
[21:15] Okay, we can do this. We can actually
[21:18] listen to loudspeakers in rooms and
[21:21] offer up opinions on those loudspeakers
[21:23] that are repeatable from person to
[21:27] person, from time to time.
[21:30] uh and uh for different kinds of program
[21:33] material. So we have a growing database
[21:37] of subjective measurements. Now we're
[21:39] going to create a parallel database of
[21:42] technical measurements so that we can
[21:44] start to stand back and compare one to
[21:46] the other and hopefully figure out what
[21:48] it is in the technical performance of
[21:51] loudspeakers that makes some appealing
[21:55] to people at large and others that are
[21:57] not.
[21:59] We have to admit right away
[22:02] that there is no single curve, no
[22:05] frequency response curve or whatever
[22:07] curve you choose to identify it as that
[22:11] can create then that can completely
[22:13] describe the performance of a
[22:14] loudspeaker. It radiates sound in all
[22:17] directions into a somewhat reflective
[22:20] space, a room.
[22:23] And the sound field by the time it gets
[22:24] to the listener is extremely
[22:27] complicated.
[22:29] There's no way to unravel that that
[22:31] complex structure. You can't just put a
[22:33] microphone at the listening location and
[22:36] and somehow analyze what what it picks
[22:39] up and go back to figure out from that
[22:43] what the loudspeaker sent out. You you
[22:45] it it's impossible. You can't go
[22:47] backwards. You can go forwards though.
[22:50] You can measure the sound that leaves
[22:52] the loudspeaker
[22:54] in all manner of different directions
[22:57] and surmise from those measurements what
[23:01] would arrive at a listener's location in
[23:03] a room. So here's what we do.
[23:06] We measure we do the spinorama that Vlab
[23:09] mentioned. It's 70 measurements at 2
[23:11] meters in an anooic chamber. We spin the
[23:13] speaker horizontally and then lay it on
[23:15] its side and spin it get a vertical
[23:17] orbit.
[23:19] 70 measurements to measure the frequency
[23:21] response. In those 70 measurements, we
[23:24] look at the direct sound, the onaxis.
[23:26] It's the first sound to arrive at the
[23:28] listener in the sweet spot. But we tend
[23:31] to want these days to entertain more
[23:33] than one person. So we do a measurement
[23:36] called the listening window, which is
[23:37] plus or - 30°, plus or - 10°, thus
[23:41] accommodating a reasonable audience.
[23:44] Frankly, that's it works at home. It
[23:46] works even here really.
[23:49] Uh, we measure early reflected sounds
[23:51] because sounds off the floor, the
[23:54] sidewalls and the ceiling are the
[23:57] strongest, the second loudest sounds you
[23:59] hear in a room are the first
[24:02] reflections.
[24:03] So, it's probably wise to pay attention
[24:05] to those.
[24:07] And to do that, what we do is we we know
[24:09] the angle that communicates that wall
[24:12] bounce to the listening area, that wall
[24:14] bounce to the listening area, the
[24:16] ceiling, and so on. So you measure those
[24:17] offaxis angles and put those into the
[24:20] equations. And to get the late
[24:22] reflections, everything else you just do
[24:24] the sound power, which is the sound
[24:26] radiated in all directions
[24:27] indiscriminately. Just add it all up.
[24:30] And the computer will chew away for a
[24:33] nancond at that data set and spit out a
[24:39] a family of curves like this. The on
[24:41] axis is the solid top line.
[24:44] Then there's the listening window. And
[24:46] ideally, you want the listening window
[24:47] to be as nearly the identical as
[24:50] possible to the on axis so that the
[24:53] person in the sweet spot and the people
[24:54] adjacent get the same kind of direct
[24:57] sound because the first sound to arrive
[24:59] carries with it the essential tambberal
[25:01] signature of the voice or the musical
[25:04] instrument.
[25:06] And then we were interest we're
[25:07] interested to see if the follow-up
[25:08] sounds match that tambberal signature
[25:11] the the reflections from the rest of the
[25:13] room whether those are communicating
[25:16] similar tambberal cues to our ears. So
[25:18] we look at the the early reflected
[25:21] sounds and these are the this is the
[25:23] energy sum of the four ceing sidewall
[25:26] bounces and back and front wall bounces.
[25:29] Then we add up in a very in a weighted
[25:32] sense. It's a complicated calculation,
[25:34] but we actually compute the estimated
[25:36] sound power radiated from the
[25:38] loudspeaker without regard to direction.
[25:42] And we then compute two more numbers.
[25:45] The directivity indexes
[25:48] uh zero in as DI directivity index of
[25:51] zero would be an omniirectional source
[25:54] sound radiated equally in all
[25:55] directions. And that is the case with
[25:57] all woofers and subwoofers. Once you're
[26:00] below a fre once the sound source is
[26:04] small compared to the wavelength it
[26:06] radiates, it radiates in equally in all
[26:09] directions. And so the wavelength at 20
[26:12] Hz is 57 ft. At 100 hertz is 11.3 ft.
[26:19] And the woofer is only a foot. So no,
[26:21] even at 100 hertz, the woofers are
[26:24] essentially omnidirectional. They don't
[26:25] care which way you point them.
[26:28] And so we calculate these directivity
[26:30] indexes as a function of frequency. And
[26:31] they are really telling us something
[26:33] about the uniformity with which the
[26:36] loudspeaker radiates its energy into the
[26:39] room. At low frequencies, it's
[26:41] omniirectional and as you go up in
[26:42] frequency, it becomes more and more
[26:44] progressively directional facing the
[26:46] audience. And then at high frequencies,
[26:49] it will be quite narrow. What we want is
[26:52] a loudspeaker that maintains something
[26:54] resembling a constant directivity over a
[26:57] wide range of frequencies. In that way,
[26:59] and only in that way, will we achieve
[27:02] this uh ideal performance of the direct
[27:06] sound matching the tambber of the
[27:09] reflected sounds.
[27:12] Uh this pinorama, by the way, is now
[27:14] embodied in a in a standard uh published
[27:17] in 2013 2013.
[27:20] And um it's uh we we'll see if anybody
[27:23] uses it. Uh the problem there are two
[27:26] problems with it. It's difficult to
[27:28] measure. Not every not every
[27:30] manufacturer has that capability.
[27:32] And even if they go outside of their own
[27:34] premises and hire a third party to do
[27:36] the measurements, which they could do,
[27:38] they might not like what they see. And
[27:41] so if they don't like what they see,
[27:42] you're not going to see it. That's
[27:44] really the fact of it. So it's quite
[27:46] it's a challenge. But the the data, if
[27:49] you can find this data, it's useful.
[27:53] Here's an example from my old original
[27:55] work done 1985, published in ' 86.
[27:59] Here's a loudspeaker that had some
[28:00] directivity problems. It was designed, I
[28:04] know the designers,
[28:06] and I know the history of it. It was
[28:08] designed by a group of engineers who
[28:10] were smart, who were told to get the
[28:12] flattest, smoothest onaxis response they
[28:15] could possibly get. And that was the
[28:17] only instruction they were given. And
[28:19] they did it. And I they went out, I'm
[28:21] sure, and to the local pub and had a few
[28:24] beers to to celebrate when they did it.
[28:27] But what they ignored because they
[28:30] weren't told to pay any attention to it
[28:32] was what happens offaxis. So we get this
[28:35] pristine first sound arriving at the
[28:37] sweet spot. But 60 30 to 60 degrees off
[28:41] axis. No, not so not so good. So, what
[28:44] does this mean in terms of how it
[28:47] sounds? Well, in terms of how it sounds,
[28:49] I can tell you that when we put this one
[28:51] into a double blind listening test, it
[28:53] didn't win and it didn't lose. It was a
[28:55] good solid mediocre loudspeaker,
[28:58] which was and it was expensive, high-end
[29:01] loudspeaker. It was a K 1052 for those
[29:04] of you who might have gray hairs and can
[29:06] remember
[29:08] um which was highly regarded at the
[29:10] time.
[29:12] So here is the direct sound that arrives
[29:15] from that loudspeaker at a listener.
[29:18] There are the that's the energy sum of
[29:20] the first four reflections. Only the
[29:22] first four reflections floor, ceiling,
[29:24] sidewalls. And already we can see in the
[29:27] measurements why that loudspeaker had a
[29:30] personality, why it wasn't neutral,
[29:34] why it it just sounded not quite right.
[29:37] We we didn't give it our highest rating.
[29:39] We didn't give our lowest rating because
[29:40] there are worse loudspeakers than that.
[29:43] But it's not perfect by any means. And
[29:45] when you go to the sound power, the
[29:47] total sound radiated in all directions,
[29:49] you get that. And you can see that in
[29:51] the total sound power and in the early
[29:53] reflections, this inferior offaxis
[29:57] performance is the dominating factor.
[30:00] The fact that it's beautifully smooth
[30:01] and flat on axis is is not a factor at
[30:04] all.
[30:05] In fact, you think about this. You look
[30:07] at that set of data and you ask the
[30:09] question, if we are to look at a data
[30:11] set and try to predict how a loudspeaker
[30:14] will sound in a room, what do we have to
[30:16] measure? Well, at low frequencies, we
[30:19] got to measure sound power. That's
[30:22] that's 10 dB higher than anything else.
[30:25] At highest frequencies, we can rely on
[30:27] the on axis because everything else has
[30:29] been attenuated. The on- axis is the
[30:31] dominant factor at high frequencies
[30:33] because the tweeter's directional and
[30:35] the room is absorptive and air is
[30:37] absorptive.
[30:39] But once you get below about 10,000 10
[30:41] kHz, everything gets into the act. You
[30:44] can ignore absolutely nothing and the
[30:45] on-axis response is not the dominant
[30:47] factor. So let's add those up
[30:51] and we get a room curve.
[30:54] Now what we see in that predicted room
[30:57] curve is evidence that
[31:00] uh as a function of time something is
[31:02] happening. The direct sound the first
[31:04] sound to arrive at your ears will be
[31:06] this black curve right here the black
[31:08] one. But as a function of time as those
[31:12] reflected sounds arrive at your ears
[31:16] they're going to add up to that red
[31:19] curve at the top the prediction. So now
[31:22] we see that what we measure in a room is
[31:25] dominated by the off-axis performance of
[31:27] the loudspeaker
[31:29] and the room itself because the room has
[31:32] to communicate that energy to us and if
[31:34] it doesn't communicate that energy
[31:36] equally at all frequencies it can't be
[31:39] neutral.
[31:41] So we have two issues here. That's an
[31:43] interesting test. Let's just say let's
[31:45] just see how that pred prediction works.
[31:48] So, we put that that loudspeaker into a
[31:50] listening arm,
[31:53] put it down in a typical stereo left
[31:55] location, measure it. Averaged, it's
[31:58] averaged over those six seats.
[32:00] Averaged over the six seats. Move it. We
[32:04] only moved it half a meter, 18 in. Not
[32:06] very far, but look at the difference in
[32:08] the curves. Wow. You either had a kick
[32:12] drum that's a little fat or a kick drum
[32:14] that was almost absent.
[32:16] It's not subtle. That's that's what
[32:18] standing waves in a room do at low
[32:20] frequencies.
[32:21] And if you play around with the
[32:23] positions, you can find another position
[32:24] that looks pretty good. And if you
[32:26] listen while you're doing all of this,
[32:28] you find, yeah, that's the one that
[32:29] actually sounds like pretty tight. Good
[32:32] kick drum. I like that. Bass notes are
[32:34] all there. Uh, and there's our
[32:36] prediction.
[32:39] So, what do we conclude? Well, first of
[32:41] all, the prediction worked. Not at all
[32:44] frequencies, but above three or 400
[32:46] hertz, I'd say we nailed it.
[32:50] So, there's enough information in the
[32:52] anacoic data of that loudspeaker to be
[32:54] able to predict
[32:56] what's happening in a room above two
[33:00] above three or 400 htz. What happens at
[33:01] lower frequencies? Well, those are the
[33:04] very familiar room resonances, the
[33:06] standing waves, the modes, the iigen
[33:09] modes. People have many names for them.
[33:11] They're the corruption of the base in
[33:13] all small rooms of any shape. There is
[33:17] no escape from them. So you have to deal
[33:19] with them. And I'll if I have time I'll
[33:21] show you how. So the prediction of that
[33:24] room curve is that tells us that the
[33:27] shape of the room curve and I'd have to
[33:29] say probably what we hear
[33:34] substantially is dictated by what
[33:36] happens offaxis. It's reflected energy.
[33:40] The only way to get rid of that
[33:41] reflected energy is to listen outdoors
[33:44] or in a very anooic room. Neither of
[33:47] which is very pleasant for listening to
[33:49] music. Right? Rooms rooms are good for
[33:52] music. Even good rooms for loudspeaker
[33:55] reproduction of music. And in this case,
[33:58] if if you do a room measurement and see
[34:00] that red curve up there
[34:03] and you reach for your trusty equalizer
[34:06] and smooth it out, what you've done is
[34:09] you've screwed up the only good thing
[34:11] that speaker had, which was its onaxis
[34:13] response.
[34:15] You have you've not solved the problem.
[34:16] You can't solve that problem with an
[34:18] equalizer. The only solution for that
[34:21] problem is a loudspeaker that behaves
[34:23] itself. So you take that one back to the
[34:25] shop and buy a better loudspeaker
[34:28] like that one. That's a better
[34:30] loudspeaker. Smooth on axis and quite
[34:33] well behaved offaxis
[34:36] and that's not a bad directivity index
[34:39] is you know it's not perfect but it's
[34:41] not horrendous. You measure put it in a
[34:43] room and you measure 6C average you get
[34:46] that
[34:47] uh there's the sound power
[34:50] and the inverse of the directivity
[34:54] index. If you turn the directivity index
[34:56] upside down you will get an estimate of
[34:58] the sound power if that loudspeaker is
[35:03] equalized to be perfectly flat or
[35:05] created to be perfectly flat on axis.
[35:07] Now, this one was almost perfectly flat
[35:10] on axis to begin with. And so, the
[35:12] inverse of the directivity index and the
[35:15] sound power that we actually measured
[35:17] sit one on top of the other. And if we
[35:20] sit move them down, they give us a
[35:23] pretty good picture of the room curve
[35:26] above 200 hertz. Only when you get up
[35:29] into the, you know, above seven or eight
[35:31] kilhertz does the tweeter, does the
[35:33] tweeter start to dominate and the sound
[35:36] power curve doesn't weight the tweeter
[35:37] tweeter output sufficiently. So, we have
[35:39] to we have to add some some more
[35:42] information. But I can tell you that
[35:44] what we're seeing is that over most of
[35:46] the critical part of the frequency
[35:48] range, we can predict a room curve from
[35:50] the anooic measurements. It's not a
[35:52] mystery. There's another one. Now, this
[35:54] is not a perfect loudspeaker. This is a
[35:57] modern version of what I just showed
[36:00] you. In fact, I heard a pair of those
[36:02] today. They're here. And the on axis is
[36:06] not flat and smooth and the offaxis
[36:08] certainly isn't flat and smooth. The
[36:11] directivity index is not not smooth. But
[36:15] because the on- axis is not smooth, then
[36:17] we will see that these curves are not
[36:19] the same.
[36:21] And let's just see that sound power now
[36:23] should be the best predictor of the room
[36:25] curve. And by golly, it is. And so then
[36:28] and the uh inverse of the directivity
[36:30] index is not bad. It's if you had
[36:32] nothing else to to work with, it would
[36:34] be that it would be pretty good. All I'm
[36:37] getting hoping I'm developing in your
[36:39] minds is the idea that these cold
[36:42] scientific numbers and graphs that we
[36:45] measure in our anacoic chambers are
[36:47] quite relevant to what we hear in rooms.
[36:50] They're not abstract.
[36:52] Now, that loudspeaker has two problems.
[36:54] It's not flat on axis and an equalizer
[36:57] can fix that because an equalizer just
[36:59] changes the amplitude response the
[37:01] frequency response but it has
[37:04] problematic directivity and an equalizer
[37:07] cannot fix that and both of those create
[37:11] contribute to the fact that the room
[37:12] curve is not flat. If you don't know
[37:15] that top data about the loudspeaker and
[37:19] all you do is measure the room curve,
[37:21] which most people do, your auto a lot of
[37:24] receivers will do this automatically for
[37:26] you. You press the button and it makes
[37:28] funny sounds and reassures you that it
[37:31] has calibrated your loudspeaker and room
[37:34] combination. Well, that's what it's
[37:36] measuring. And it has no idea what's
[37:38] wrong with the loudspeaker or what's
[37:40] wrong with the room. Doesn't know what's
[37:42] causing it. So it doesn't know what to
[37:43] do to correct it. In this case,
[37:45] equalization is not going to solve the
[37:47] problem. Only a better loudspeaker will.
[37:51] Now these are popular terms. I don't
[37:53] know. Does anybody here have a room
[37:56] equalizer in their home system or
[37:58] business system?
[38:00] Yeah, it had to be one. Thank you.
[38:04] No, because the if you go out to your
[38:06] local if you order buy any of the
[38:08] modernday uh uh domestic uh surround
[38:13] surround sounds receivers, many of them
[38:16] have these algorithms built into them.
[38:19] Uh you can go out and spend hundreds or
[38:21] thousand or more dollars for a
[38:23] standalone uh box that does this for you
[38:26] or you can pay somebody to come in and
[38:28] do it for you. And and it's it's a
[38:30] system. It's an or it's a service and
[38:33] the claim is that by making such
[38:35] measurements on unknown loudspeakers in
[38:37] unknown rooms that we can create
[38:40] perfection for anybody anywhere. Uh and
[38:44] unfortunately there are some standards
[38:46] from simp the motion picture industry
[38:49] ISO ITU that actually promote this
[38:53] belief and I'm afraid to say it is
[38:55] simply not true. it can't be true and it
[38:58] isn't true and in my dying years I'm
[39:03] going to do my damnedest to to change
[39:05] it. I mean the fact is that when you
[39:08] consider all of the audible
[39:09] characteristics that two ears in the
[39:11] brain can extract from the sound field
[39:14] of sound source in a room, it's not
[39:17] going to be accurately represented by a
[39:19] microphone and a simple analyzer. It's
[39:21] just it's a presumption that isn't true.
[39:25] what we're measuring is not what we're
[39:26] hearing. So back to now what we are
[39:29] measuring
[39:30] and what we are measuring is is that uh
[39:34] the loudspeaker is in control over part
[39:36] of the frequency range. The room is in
[39:37] control over part of the frequency range
[39:39] and the transition frequency between the
[39:42] room and the loudspeaker control varies
[39:44] with the size of the room. And as the
[39:48] room gets larger, the loudspeaker takes
[39:51] over more and more control. So in a room
[39:53] this size, the loudspeaker is going to
[39:56] be in control down to probably oh 40 50
[39:59] hertz.
[40:01] And above and below that, we're going to
[40:03] have to pay some attention to this
[40:04] particular space. But above that, we
[40:07] have some assurance that what you're
[40:09] going to hear will be dominantly
[40:12] influenced by the loudspeaker. And so in
[40:15] theory, if I'm right, all I need to do
[40:17] is to put a really good loudspeaker
[40:19] here, turn it on, and you'll be a happy
[40:22] camper. Uh, except at low frequencies.
[40:25] We may have to pay some attention to low
[40:26] frequencies. Now, the interesting thing
[40:28] is that I was thinking along these
[40:30] lines, and then I stumbled into man
[40:33] named Art Benade,
[40:35] a physicist, musician,
[40:37] brilliant man, dead now. And but before
[40:40] he did, he died. He sent to me a
[40:42] collection of his writings
[40:45] and this was among them and I said well
[40:47] this is it
[40:49] because he just said well look you know
[40:50] as a musician it's obvious the musical
[40:53] source is well defined I mean I can play
[40:55] the same riff 10 times and it'll sound
[40:57] it'll be the same acoustical output 10
[41:00] times and go back in the audience it
[41:02] sounds the same 10 times and but if you
[41:05] make measurements and you walk around
[41:06] the audience and it still sounds the
[41:08] same but if you make measurements from
[41:10] here to there it's going to be very
[41:12] different. You put the microphone in
[41:14] different positions you make you find
[41:16] huge huge huge variations
[41:20] he's suggesting or my interpretation of
[41:22] his suggestion is that if we were if
[41:25] we're able to describe the m the sound
[41:27] source in measurements
[41:30] then we should be able to predict what
[41:33] listeners are hearing. what happens in
[41:35] between maybe doesn't matter quite so
[41:38] much because they're going to be very
[41:40] very difficult to interpret.
[41:43] I mean, if you make measurements in a
[41:44] concert hall, you know, put a sound
[41:45] source on the stage and measure back in
[41:47] the audience and do a, you know,
[41:49] frequency response transfer function,
[41:51] what have you, it's going to be awful,
[41:54] awful, awful, awful. An engineer will
[41:56] look at it and say, "This is just a
[41:58] staggeringly complicated comb filter.
[42:00] Can't possibly
[42:02] sound good. communication may be
[42:04] impossible and yet we go there and we
[42:07] listen to it and we love it. It's a
[42:08] reference listening experience. That's
[42:11] the difference between what what we are
[42:13] measuring and what we are hearing and in
[42:15] this case we are measuring the wrong
[42:17] thing. So let's just draw the parallel
[42:19] between live performance and
[42:21] reproduction.
[42:23] We have musical instrument plus room to
[42:26] the brain. Brain sorts it all out. Now,
[42:29] if we have a good recording of that mu
[42:31] that instrument reproduced by a good
[42:33] loudspeaker, we should be able to sort
[42:34] of have a parallel experience in our in
[42:37] our homes.
[42:39] And that's that's the challenge. That's
[42:40] what we're going to see if it happens.
[42:41] I'll tell you it does happen actually.
[42:44] But the success of that whole exercise
[42:46] will depend on the competence of the
[42:48] loudspeaker designer and
[42:50] on the ability of the recording
[42:52] industry. That's you people, some of you
[42:54] anyway, to capture and store accurate
[42:59] renderings of what was the sounds that
[43:02] were radiated by those voices and
[43:04] musical instruments that we care to
[43:06] listen to.
[43:08] And this is an artistic exercise. And uh
[43:10] you just go into any one of these
[43:11] control rooms and you see these massive
[43:13] consoles and offside apparatus with
[43:16] knobs and dials and all kinds of
[43:18] processing available and that is
[43:20] perfectly legitimate.
[43:22] But you know we it it is it's legitimate
[43:25] and it's part of the problem. Not that
[43:29] the art shouldn't be whatever the artist
[43:31] wants it to be. It's just that we have
[43:33] trouble delivering that art to the cons
[43:35] to the customer.
[43:37] See, we listen in homes, in our cars,
[43:38] and through headphones. That's the audio
[43:40] industry. Audio industry is you capture
[43:42] sound, you process it, you send it off
[43:44] to a live audience in real time. You can
[43:47] store it, you can send it off to
[43:49] customers in their wherever they want to
[43:51] press a play button. Or you can send it
[43:54] off to a different form of distort of
[43:56] storage and send it off to the cinema
[43:58] world. So that's the audio industry as I
[44:01] know it, as I've worked in it all of my
[44:04] professional career.
[44:06] And my philosophy is embodied in this.
[44:10] Science in the service of art is our
[44:12] business. And when you combine those
[44:15] three words, it's kind of oxymoronic.
[44:17] Science and art together. Yeah. What's
[44:20] business? Science and business. No, it
[44:22] doesn't work. Art and business in music.
[44:25] Yes. But good sound is our product. I
[44:28] mean, I worked for a manufacturer that
[44:29] made boxes that cost money, wooden
[44:32] boxes, steel boxes. But it is it's the
[44:35] sound that's the product. That's what
[44:37] puts the smile on the customer's face or
[44:39] the frown on the customer's face. And
[44:41] you know, if we screw up, the customer
[44:43] is not happy. And it's not happy because
[44:45] the box doesn't look good. He's not
[44:47] happy because the sound isn't good. So,
[44:50] how do we know what good is? We listen.
[44:52] And when we listen, we get caught into
[44:53] what I call the circle of confusion
[44:55] where loudspeakers
[44:57] are evaluated using recordings.
[45:00] And those recordings were created in
[45:02] control rooms where you play with
[45:03] microphones, select microphones,
[45:05] position microphones, play with
[45:06] equalizers, add spatial temporal effects
[45:09] and what have you while you are
[45:12] monitoring what you're doing through
[45:13] loudspeakers.
[45:15] And so you know where this is going,
[45:16] right? I mean the loudspeakers aren't
[45:18] standardized.
[45:20] uh over the years they have varied
[45:23] horrendously
[45:26] and I've recently even been into control
[45:28] rooms where I wouldn't give house room
[45:31] to the speakers that they are listening
[45:33] to. They're creating art through really
[45:36] dreadful products, dreadful
[45:37] loudspeakers.
[45:39] And as a result, the recordings are
[45:40] variable. I mean, I was talking to this
[45:44] just not long ago saying, you know, I
[45:46] sat down with headphones on, which I
[45:47] don't normally do very much. I don't
[45:49] really care much for headphone
[45:50] listening, but in this case, I was going
[45:52] to explore the streaming music sources
[45:55] and and see how they were doing for
[45:58] keeping the sound levels, the playback
[46:00] levels relatively constant from tune to
[46:02] tune. That's they do that quite well
[46:04] sometimes. But they also I realized
[46:07] immediately that the the the tambberal
[46:09] balance the spectral balance was varying
[46:12] hugely. So, I mean, I went into the the
[46:15] algorithm and played with the bass
[46:16] control, played with the treble control,
[46:18] and and found that really quite a high
[46:20] percentage of the of the recordings I
[46:22] was listening to benefited from tone
[46:24] control adjustments. Those oldfashioned
[46:27] things, bass and treble that we my
[46:30] daddy's
[46:31] apparatus used to have, you know, and
[46:34] it's because they are variable. Well,
[46:36] why are they variable? Part of it at
[46:38] least has to be the monitoring
[46:40] circumstances because if the monitor
[46:42] speakers are radiating too much bass,
[46:44] you don't put as much bass into the into
[46:46] the disc into the into the recording and
[46:48] on and on it goes. I mean, you can see
[46:50] where this is not very it's not heavy
[46:52] medicine here. We have to have some
[46:55] relationship between the domain within
[46:57] which the art is created and the domain
[47:01] within which the art is appreciated.
[47:03] There has these professional monitor
[47:05] loudspeakers in rooms have to behave in
[47:07] some ways similar to our consumer
[47:10] loudspeakers and rooms or cars even or
[47:13] the system doesn't work. The art that is
[47:16] being so carefully created by the
[47:18] performers and by the recording
[47:20] engineers that art is not being
[47:23] communicated to the audience in a way
[47:26] that allows the audience to say with any
[47:28] certainty that if I don't like it I
[47:32] don't know who to blame. Do I blame my
[47:33] loudspeakers? Is it my room that's the
[47:36] problem or was it just a crappy
[47:38] recording? That's that's the dilemma
[47:40] you're in. Unless you have measurements.
[47:44] And so the we can anticipate the sound
[47:48] field in a room as the physical sound
[47:51] field in a room from that spinorama
[47:53] data. The real challenge now is does
[47:55] that translate into uh what I what we
[47:58] perceive.
[48:00] And so, uh,
[48:02] one day I get a phone call from the CEO
[48:06] of Harmon
[48:08] saying,
[48:09] "Tool, we pay you a lot of money to make
[48:13] sure our products sound good." Yes.
[48:17] Well, can you explain the recent or the
[48:19] current uh Consumer Reports review of
[48:22] loudspeakers in which some of our
[48:24] products did not do well?
[48:26] Well, I I said, "Can I talk?" So, I went
[48:28] into his office and we talked and I
[48:31] explained to him that for decades, those
[48:34] of us in the in the loudspeaker business
[48:37] knew that Consumer Reports reviews
[48:39] didn't really mean anything because they
[48:42] were doing they were doing it wrong.
[48:44] Well, he said, "Can we teach them to do
[48:46] it right?" And I said, "Yeah, probably.
[48:49] It's going to take some money and some
[48:50] time." And so, he authorized it. He just
[48:52] says, "Okay, do it. Make it make it so."
[48:55] So Sean Olive and I got together and we
[48:57] put together
[48:58] what we knew about the psycho acoustic
[49:01] relationships between these curves and
[49:03] numbers and and subjective opinions.
[49:07] And we went back to our database because
[49:10] at Harmon we do these listening tests
[49:13] routinely on and we do the measurements
[49:15] on our own products and on our
[49:17] competitor's products so we know where
[49:19] we stand. And at that particular time we
[49:22] had a database of 70 on 70 loudspeakers.
[49:26] We had div these double these double
[49:28] blind single listener meticulous
[49:31] listening tests where the result was a
[49:33] score on a scale of 10. Now that's
[49:36] listening to the loudspeaker in a room.
[49:38] So that's the combination of the
[49:39] loudspeaker and the room that's being
[49:41] rated.
[49:43] Part two of the test is to take those
[49:44] loudspeakers into the anacoic chamber,
[49:48] do the spinorama,
[49:50] and process that spinama data through an
[49:54] algorithm that tries to pay attention to
[49:56] spectral balance on and off axis and
[50:00] bandwidth
[50:02] um narrow band tambberal problems,
[50:05] resonances and what have you
[50:11] and come up with a number. So we have a
[50:13] calculated number based on the anooic
[50:15] data and we have the number that came
[50:18] out of the listening test and the
[50:19] correlation was 086.
[50:21] That that's not chabby. You go to Las
[50:24] Vegas with those odds, they'll kick you
[50:26] out of the casino in no time.
[50:29] Uh but going back to the origin of the
[50:32] that the motivation for this consumer
[50:34] reports. So there oh yes there's the
[50:36] pretty perspective on it from meic data.
[50:39] actually predicted listener repont re
[50:42] reactions to the loudspeaker.
[50:45] But we did 70 loudspeakers. And in that
[50:48] 70 loudspeaker collection, there were
[50:50] big expensive speakers and there were
[50:52] little inexpensive speakers that had no
[50:55] bass and had no pretense to excellence
[50:57] perhaps, but could be quite good over a
[51:00] smaller bandwidth, narrow bandwidth.
[51:03] Uh so we separated out from that 70 13
[51:09] the original 13 that Consumers Union had
[51:11] published and there's the the result on
[51:15] the left. This these big dots here are
[51:18] the results of the listening tests our
[51:20] double blind double blind listening
[51:22] tests from the worst down here up to the
[51:26] best up there. The square, the open
[51:28] squares are the Consumer Reports
[51:30] ratings.
[51:32] And you can see our best is their worst.
[51:36] Uh, and this is this is fascinating. It
[51:38] was a slightly negative correlation,
[51:42] but with a very low probability. So
[51:45] there's a lot of variation.
[51:47] And when we put it through uh the
[51:51] algorithm calculating algorithm that
[51:53] Sean had created uh the correlation was
[51:56] 0.995. Well that's one that's perfect.
[51:59] The calculated subjective ratings and
[52:02] the real subjective ratings just agreed.
[52:05] I see them right on the same line.
[52:07] Here's a very specific example kind of
[52:10] an interesting one. There are four
[52:11] loudspeakers. These were all high-end
[52:13] audio file approved,
[52:16] highly regarded uh reviewed uh
[52:19] loudspeakers.
[52:21] If you followed these magazines and read
[52:23] these reviews, you would feel quite
[52:25] confident in going out and if you had
[52:28] something of the order of $10,000 to
[52:30] spend, which is what these cost each
[52:34] pair. Um, if you had that kind of money
[52:37] to spend, you would feel quite confident
[52:39] that if you bought any one of them,
[52:40] you'd be buying something resembling the
[52:42] state-of-the-art.
[52:44] But in our listening tests, they weren't
[52:47] all state-of-the-art. There were two
[52:48] that were in a close horse race for
[52:50] first place, and the other two were just
[52:52] further further down. And that's the
[52:54] spinorama data.
[52:56] And as they get flatter and smoother,
[53:00] the scores get higher. It's not magic.
[53:04] We would never ever buy an electronic
[53:07] component
[53:09] and nobody would ever design an
[53:10] electronic component that didn't have a
[53:12] frequency response that was flat from,
[53:15] you know, 10 hertz to 100 kHz, you know,
[53:18] that kind of thing. It's flat and
[53:20] smooth. It's not not difficult to do and
[53:23] everybody does it. Why would we think
[53:25] that when you get to a loudspeaker, we
[53:27] can kind of throw that rule away? Well,
[53:29] it turns out that what's what our
[53:31] listeners here are doing is saying, "No,
[53:33] you shouldn't throw that rule away. It's
[53:34] a good rule. Do it. And if you do it,
[53:37] your product's going to sound better. I
[53:39] will like it more."
[53:41] Uh,
[53:44] looking at the there were 350 listeners
[53:46] that participated in that test over 3
[53:49] month or so period. Uh, 268 had done the
[53:53] test when Sean wrote his paper. It's in
[53:55] the journal of the AES.
[53:57] uh showing how the listener's
[54:01] performance of experience and background
[54:03] related to their performance and their
[54:05] performance here was based on the FL
[54:08] statistic. It's a measure of how
[54:10] consistent people are when they
[54:12] experience the same loudspeaker on
[54:13] different occasions because we do
[54:15] multiple time we we hear the they hear
[54:17] the speaker several times in different
[54:19] orders, different sequences
[54:22] and on the strength of their their
[54:25] differentiation. If they hear some
[54:27] loudspeakers that are from good to bad
[54:29] and they you know give the good ones
[54:31] sixes and sevens and the bad ones twos
[54:33] and threes, we know that's strong
[54:34] differentiation. But there are people
[54:36] who don't have don't do that. they will
[54:38] give you sixes and sevens and fives and
[54:40] sixes. They they don't really spread it
[54:43] out. It turns out that the ones that
[54:46] exist exhibited both of these
[54:48] characteristics most are are selected
[54:51] and trained listeners. They first of all
[54:53] is they're selected on the basis of
[54:55] having normal hearing. Uh not everybody
[54:58] has normal hearing. So that's the first
[55:00] hurdle they have to pass and then they
[55:02] are actually trained to recognize
[55:04] different kinds of defects.
[55:07] Uh the retail salespeople did quite
[55:09] well. The audio reviewers, six of them
[55:12] who volunteered to come in, they were in
[55:14] they were bold people. They didn't do
[55:16] great. The our own brand and market
[55:19] sales and marketing people did really
[55:21] poorly.
[55:24] These are the people are going out into
[55:25] the field say buy our product. It's
[55:27] great. I you can v I'll vote for it.
[55:33] Please don't.
[55:35] And then the students, these these were
[55:37] not students like you people in a music
[55:39] and sound recording environment. These
[55:41] were just generic students. Uh
[55:47] non-auudio non-m music students.
[55:51] Anyway, it's kind of interesting. Point
[55:53] is though look they very they they were
[55:56] differ they differed in their ability to
[55:59] give consistent answers and they
[56:01] differed in their abilities to give
[56:03] numerically spread out answers but they
[56:06] in the end they all agreed on the
[56:08] ratings of those products. There were no
[56:12] no inconsistencies. It was really humans
[56:16] are amazing measuring instruments.
[56:19] Amazing. I mean, I've done this for 40
[56:21] years, and I still kind of say, "Wow,
[56:26] this this is remarkable."
[56:28] Because, as I say, you're listening to
[56:30] strange music in a strange room. It's a
[56:33] strange loudspeakers, and you're forming
[56:35] opinions. And the only way we can get
[56:38] those consistent opinions out of you is
[56:39] to not let you know what you're
[56:40] listening to. Do it blind. But if you
[56:43] know what you're listening to, I don't
[56:44] care what you think. You, it doesn't
[56:46] matter.
[56:48] And so, you know, I go back to to Art
[56:50] Bened's
[56:53] idea. He was absolutely right.
[56:56] Absolutely right. The difference is that
[56:58] we can measure loudspeakers that are
[57:01] reproducing musical instruments. We
[57:04] don't have any parallel way to measure
[57:05] the musical instruments themselves
[57:07] because they are so different and so
[57:09] varied and, you know, so peculiar in the
[57:11] way they radiate sound into rooms. So we
[57:14] have those two now measurements those
[57:16] two measurements now and we we we
[57:17] combine them and that gives us some
[57:20] psycho acoustic relationships and what
[57:22] do we learn from that is that to make a
[57:25] good loudspeaker it has to be smooth and
[57:27] flat on axis if you don't people will
[57:31] complain
[57:32] the directivity index should be smooth
[57:34] and relatively cons relatively constant
[57:37] it can change but it should change
[57:38] gradually and smoothly it seems that's
[57:40] we're still in the learning process
[57:42] there but that looks It's pretty good.
[57:45] We don't want any resonances. We've got
[57:48] to find in our measurements ways to
[57:50] identify and get rid of measure of
[57:52] resonances. And we can do that in our
[57:54] spinas. And the reason why resonances
[57:57] are so important is because they are the
[57:59] building blocks of every sound that we
[58:01] care to listen to. Human voices and
[58:03] sound of my voice is a bunch of
[58:04] resonances. Musical instruments just
[58:06] bunches of resonances. And so if you
[58:09] find additional resonances in
[58:11] loudspeakers, you're coloring the sound
[58:13] and and people recognize that quite
[58:15] quickly because that same coloration
[58:18] shows up in every recording, every
[58:20] music, every voice you play.
[58:23] So we want to get rid of that. So good
[58:24] loudspeakers are flat, relatively
[58:27] constant in their radiation patterns
[58:29] over the frequency range. They don't
[58:31] have resonances. We have in order to do
[58:33] that, we have to measure at high
[58:35] resolution. The old idea that third
[58:37] octave data is useful. I'm afraid if you
[58:40] have a third octave analyzer, give it to
[58:42] your kids because it's a toy.
[58:44] Um,
[58:46] and base performance, this is this is a
[58:48] big deal. Interesting. Base performance
[58:51] all by its lonesome accounts for about
[58:53] 30% of the factor waiting in the in the
[58:56] subjective evaluations.
[58:58] And if you don't get the base right,
[59:01] it's everything else kind of suffers.
[59:03] Getting the bass right is is is a bit of
[59:06] a trick. When you think of loudness,
[59:09] loudness in the middle frequencies
[59:10] changes by a factor of two for each 10
[59:15] decibel increment. You know, middle
[59:17] frequencies you from two or 30 hundred
[59:19] hertz to three or four kHz roughly. Uh
[59:22] you if you want if you turn the volume
[59:24] up 10 dB, most people would say yeah,
[59:26] it's about twice as loud. You turn it
[59:28] down, it's half as loud. That's a factor
[59:30] of two 10 dB.
[59:32] um except at low frequencies because the
[59:35] low the lou equal loudness contours you
[59:38] if you go back to your textbooks equal
[59:40] loudness contours crowd together at low
[59:42] frequencies and so at 50 hertz it may
[59:46] only require four or five dB to produce
[59:48] a factor of two difference in subjective
[59:50] loudness and so if the bass is is off by
[59:54] a little bit it reflects badly on the
[59:57] rest of the whole spectrum. So getting
[59:59] it right means getting it right. Um, can
[60:03] we predict any of this from manufacturer
[60:05] specifications
[60:07] because it'd be kind of nice if we
[60:09] could. I mean, it's fine for me to stand
[60:11] up here and tell you that if you have a
[60:14] spinama, you can find out all kinds of
[60:16] useful things. Predict how it sounds,
[60:18] predict how it measures. Very cool
[60:19] stuff. But you go shopping for
[60:22] loudspeakers, you don't see spinas.
[60:25] There's that's going to change slowly,
[60:27] but it right now that's the case. You
[60:29] find a few. There are some of the Harmon
[60:31] products that are have spinas. You can
[60:32] go on the website and find them. The M2s
[60:35] that you guys just bought have their
[60:37] spinas up on the website. Anyway, for
[60:40] $200 a pair, you don't expect a lot. You
[60:43] go to your local box store on a weekend
[60:45] and they have it on sale for $150 a pair
[60:48] and you put a pair under your arm and go
[60:50] home. And all they're telling you is
[60:52] that the performance of that loudspeaker
[60:54] from some 58 hertz here to 22 kHz up
[60:58] here, it's going to fall somewhere in
[60:59] that tolerance range. Plus or - 3 dB.
[61:02] Six decibel tolerant range. Six dibels.
[61:04] Gee, how many dB to double the loudness?
[61:07] Did I say 10? Well, six is Yeah, it's
[61:10] not that far away, is it? Now, if you
[61:12] see a curve,
[61:14] it shows you a lot more information. You
[61:17] get a lot more confidence looking at
[61:19] that curve. is tilted and you can bring
[61:22] that down with a tone control tilt.
[61:23] Bring flatten it out if you wanted to.
[61:25] It' be easy to do. If you knew that
[61:26] curve, if you had that curve, you could
[61:28] think about doing that. But if you don't
[61:31] have that curve, you you can't think
[61:32] about doing anything because the
[61:34] manufacturer has not told you anything
[61:35] useful.
[61:37] If you have the spinorama, wow, whole
[61:41] different ballgame.
[61:43] Now, we can see that it has resonances.
[61:45] How do I know those are resonances?
[61:46] because they are present in all of the
[61:48] curves. All the spatially average curves
[61:51] show the same little bumps and those are
[61:54] resonances and loudspeaker transducers
[61:57] are minimum phase devices. Now that is
[61:59] really technical jargon but I'll quickly
[62:02] explain. It means basically that if if
[62:04] the frequency response the amplitude
[62:06] response is smooth and flat or smooth no
[62:10] bumps then there are no resonances. If
[62:12] you see a bump there will be a
[62:14] resonance. it will ring. And if you have
[62:18] an equalizer that you can match to those
[62:20] bumps and get rid of the bumps, then the
[62:23] resonance is gone. There's no more
[62:24] ringing. You've corrected the problem.
[62:26] So, this is one of the reasons why
[62:30] active loudspeakers with built-in
[62:32] amplifiers and digital electronics
[62:35] are, I would hope, the future because
[62:38] that's the only way we can really make
[62:40] super good loudspeakers. The M2 is one
[62:43] of those. It's a It's a dedicated
[62:45] electronics.
[62:48] $1,800 a pair. Looks pretty good. Better
[62:51] than the spec actually. But here's a
[62:53] case where the marketing department
[62:55] said, "Ah, plus or - 3dB industry
[62:57] standard. Everybody understands it." And
[62:59] it's a giveaway statement because it
[63:01] didn't it does a disservice to this
[63:04] extremely good loudspeaker. So for
[63:06] $1,800 a pair, if you were fortunate
[63:09] enough to go out and find those and buy
[63:11] them, you ended up at home with
[63:13] something that is dangerously close to
[63:16] the best that can be done. Period. Then
[63:19] certainly dangerously close. So if you
[63:22] were to pick ask me what the point of
[63:24] diminishing returns would be that well
[63:26] clearly that's an example of it because
[63:28] honest to God you could pay 10 times
[63:30] that or more and not get as much in
[63:35] terms of tambberal excellence and
[63:37] accuracy as that portrays.
[63:40] Here's the 11,000 to pair. Whoa. Whoopsy
[63:43] doozy.
[63:45] Okay, so that's not great. If we could
[63:48] cheat and smooth it out and it would fit
[63:49] in there maybe. But uh we don't like to
[63:51] do that. We show the whole spinama. And
[63:54] what we see in the spinorama is that we
[63:57] have resonances folks. Now this just out
[64:00] of curio interest is this is a a a large
[64:03] electrostatic panel sitting on a
[64:05] standard woofer.
[64:07] Looks cool. Very sexy. Very industrial
[64:10] good industrial design. People just fall
[64:12] in love with it and fall all over
[64:13] themselves. Oh. And then the technical
[64:16] stories come out. Well, it's it's an
[64:18] electrostatic loudspeaker and therefore
[64:19] the mass the diaphragm has no mass. That
[64:22] means it moves instantaneously.
[64:25] It's speed. It's fast. It's accurate.
[64:29] Well, no, it's not. It's not fast. It's
[64:32] not accurate. And it rings like kick at
[64:34] a whole bunch of frequencies because
[64:37] it's very directional. Oh, yes. By the
[64:38] if you look at the difference between
[64:40] the black curve and the green curve,
[64:41] that's the on axis. That's the listening
[64:43] window. That means that no two people
[64:44] even on a sofa will hear the same sound.
[64:49] It is truly a bad design.
[64:54] And and it's a poor investment obviously
[64:56] at 11 grand a pair. Now you can go up to
[64:58] 24,000.
[65:00] Whoa. Rock and roll. Okay,
[65:03] that's that. It uses up the whole plus
[65:06] orus 3dB tolerance. You'd say, well, if
[65:08] 24 cran should have done a little more
[65:11] work than that. It meets the spec, but I
[65:14] wouldn't give it any prizes. Look at the
[65:17] spinama. Well, my goodness. What have we
[65:19] got here? We got a resonance
[65:21] for 24,000. We have a resonance that we
[65:24] didn't see in an $1,800 loudspeaker.
[65:26] Isn't that interesting? You can hear
[65:27] that cuz it's above the threshold of
[65:30] audibility. And it's not flat. It's got
[65:33] a directivity problem. A little kink in
[65:36] the directivity. It radiates widely and
[65:38] then all of a sudden it gets decides to
[65:40] get narrow.
[65:42] It's not flat anywhere. You can't
[65:44] equalize it. Uh so why bother? Why why
[65:48] would you spend that money for that
[65:49] loudspeaker? 16,000 a pair. Now here's
[65:52] this is an interesting one. It's plus or
[65:53] minus one dB. Now that's getting
[65:56] serious. That's an audacious
[65:59] challenge. And they just squeak you in.
[66:02] Just squeak it barely. Boy, that's a
[66:05] that's a brave thing to do. And I'll
[66:08] tell you why it's a brave thing to do.
[66:10] Because one of the fundamental problems
[66:13] of the loudspeaker industry is that
[66:17] maybe five I've got to quit. Uh is that
[66:19] the consistency is one thing to design a
[66:22] golden prototype and you can you know
[66:25] you might get lucky and do really good
[66:27] job on that. It's a very different thing
[66:29] to clone that in mass production.
[66:32] And uh when you know we do some
[66:36] products, this is one where we guarantee
[66:39] that it's going to be with win in spec.
[66:41] And the only way we can guarantee that
[66:42] is to spend a lot of money in the
[66:44] production process. That's part of what
[66:47] you're paying for is consistency. And
[66:49] it's sad because there are products that
[66:53] other from other sources and other
[66:56] brands what have you that you don't know
[66:58] what you're buying because they're they
[67:01] have they're small companies. They don't
[67:02] have any real production quality control
[67:05] and so on. It just it's a it's it's an
[67:08] insidious thing that does sound good by
[67:11] the way. That's that's reference quality
[67:13] sound.
[67:15] I and just my joke, there's more useful
[67:18] and reliable information on the side of
[67:20] a tire than there is in a loudspeaker
[67:23] specification sheet. And this is not
[67:26] true only of consumer loudspeakers. It
[67:29] is equally true of professional
[67:30] loudspeakers. I went on the I go on the
[67:32] websites and I checked out what their
[67:34] spec sheets are telling you. It's
[67:36] nothing. They're not telling you
[67:38] anything useful. They're saying, "Trust
[67:41] me. Trust us."
[67:44] That's the M2, by the way. And uh it's
[67:48] arguably about as good as speakers get
[67:50] these days.
[67:52] And so how does it sound? Well, depends
[67:55] on what you play through it because of
[67:58] the circle of confusion, right? And if
[68:01] if you if you have a perfect
[68:03] loudspeaker, that's pretty a good
[68:05] example of where we stand at the moment.
[68:07] And if we get even better than that in
[68:09] the future and we have somebody
[68:11] somewhere creates the perfect
[68:13] loudspeaker, is it always going to sound
[68:15] good? Hell no. Because the recordings
[68:18] are so variable. And the recordings are
[68:20] all over the place. We have no idea what
[68:22] kind of loudspeaker or room or what
[68:24] circumstances under which they were
[68:26] mixed. And so when we hear what we hear
[68:30] through
[68:32] unmeasured loudspeakers, we don't know
[68:34] what where the problem came from. we
[68:36] might like or dislike. We don't know who
[68:38] to praise. Do we praise the loudspeaker
[68:40] because it sounds good? Or if it sounds
[68:42] bad, do we condemn the loudspeaker?
[68:44] Well, it might have been the recording.
[68:47] Only if we have neutral loudspeakers in
[68:50] the production environment and neutral
[68:52] loudspeakers in the home reproduction,
[68:55] car reproduction environments. Will we
[68:58] get out of this unfortunate circle of
[69:00] confusion? Uh, we have to get people
[69:03] mixing on good speakers and people at
[69:06] home listening on good speakers. I have
[69:07] to tell you right now,
[69:09] uh, people at home are doing pretty
[69:10] well. Even cars are getting better. I
[69:13] don't know whether you've noticed. Even
[69:15] some of these little Bluetooth connected
[69:19] wireless loudspeakers don't sound half
[69:21] bad. Things are way better than they
[69:23] were 10 years ago, certainly 20 years
[69:25] ago. So, good sound is coming. And it's
[69:28] coming partly because of efforts and
[69:31] people like myself who you go out and
[69:34] find out how to make measurements and
[69:36] use them and then teach it and
[69:38] standardize it and hope people will pick
[69:41] pick up on the idea. But in the case of
[69:44] if you have good loudspeakers and you
[69:47] hear something you don't like, you know
[69:50] the problem is in the recording. It's
[69:51] not in the loudspeaker. That's the
[69:53] thing. That's the thing. I used the
[69:54] example earlier with VSLav. We were
[69:56] talking in his office and I said it's
[69:58] like that clear window and and you you
[70:01] you the recording people are creating
[70:03] the the scene outside the window. And if
[70:07] you don't like the scene, the window's
[70:08] clear. If you don't like the scene,
[70:10] change the scene. That's your job.
[70:13] That's you, the artist. But if the
[70:15] window's pink, you have no way of
[70:17] knowing. And if I have a blue window and
[70:20] you'll have you, you know, it's it's a
[70:21] silly analogy, but no, it's not a silly
[70:23] analogy because it's really it's real.
[70:25] It's the same thing when you have a
[70:27] neutral loudspeaker.
[70:29] You may not even like recordings you've
[70:31] made yourselves when you listen to them.
[70:34] But that's fine. That's the way it is.
[70:35] You you you were you weren't listening
[70:38] perhaps through a neutral loudspeaker
[70:39] when you made the mix. That's the thing.
[70:42] So from here on though, you have the
[70:44] opportunity you have the opportunity to
[70:47] use neutral loudspeakers.
[70:49] And they're at the present time they're
[70:51] getting down in price to the point where
[70:53] as I showed even 200 bucks plus a
[70:54] subwoofer gets you in the ballpark.
[70:58] Uh only then can we create art in a
[71:02] circumstance where it might be
[71:03] reproduced. And now fortunately we are
[71:06] able to measure neutrality.
[71:09] I could end there. I think I probably
[71:12] should this time. Right.
[71:15] Okay. Thank you. Actually,
[71:21] [Applause]
[71:35] I had one more slide. I have one more
[71:37] slide to show you. I I forgot I put this
[71:41] one on. This is This is the fin. This is
[71:44] the cap. This is the last last slide.
[71:46] Here's the M2 that that last
[71:51] I'm sorry.
[71:53] No, it's sorry. Okay, that's the M2
[71:56] measured in a the red curve was measured
[72:00] in one of our home theater rooms at
[72:02] Harmon. Six six seats, just a small
[72:05] typical domestic home theater.
[72:08] And the other curves are measured in
[72:10] film sound facilities ranging from 24,
[72:13] 60 seats, 114, 100 to 516 seats. You
[72:18] know, pretty big
[72:20] measurements are the same up above
[72:24] the frequency above where the
[72:26] loudspeaker takes over. Below where the
[72:27] loud where the room takes over the you
[72:29] got a different problem. You got a
[72:31] different problem at low frequencies.
[72:33] That's that's the room getting in the
[72:35] way. And we have ways of working on
[72:37] that. I can talk about them tomorrow.
[72:40] But uh above the good news is that above
[72:43] about 200 hertz, we're done. And I'll
[72:46] have to tell you that these little
[72:48] irregularities in the cinema, the black
[72:50] curves here, those ones that the stray
[72:52] ones that stray out, those were the ones
[72:55] that were measured with the microphones
[72:56] at ear level,
[72:58] which puts the microphone dangerously
[73:01] close to the back of the seat, which is
[73:02] an acoustical interference thing that
[73:04] the human two ears and a brain just they
[73:06] don't we don't hear it. You can stand up
[73:09] and sit down in a cinema sometime or in
[73:11] a concert sometime if you want to be
[73:12] impolite and and it just doesn't you
[73:15] don't hear that kind of thing but you
[73:17] measure it and if you measure it the
[73:19] temptation on the part to the te to the
[73:21] technician is that they will equalize it
[73:24] and in equalizing it they're equalizing
[73:26] something that cannot should not be
[73:28] equalized and they're making what might
[73:31] have been a really good loudspeaker
[73:32] sound worse so now I am through thanks
[73:38] [Applause]
