Why does Rob distinguish perception from hearing, and why does he distrust psychoacoustic thresholds?
Short answer
Watts says he talks about perception, not hearing: hearing is the ear's abilities, which he calls very limited, and perception is the brain turning the ear's data into separate, placed sounds, a process he says science has little understanding of Head-Fi, Sep 2018. He distrusts psychoacoustic thresholds because, as he describes them, they come from simple sine-wave or single-tone tests of the ear and say nothing about that brain processing Head-Fi, Mar 2016Head-Fi, Sep 2018RMAF 54. So, he says, an error below a threshold cannot be assumed inaudible, and careful listening tests must decide Head-Fi, Oct 2014RMAF 1:41. His proposed mechanism is the brain's use of correlation, which he usually hedges ("probably", "my guess", "I suspect") but in an October 2025 podcast (published March 2026) called "definitely" the case for noise floor modulation Head-Fi, Mar 2016MFG Audio Reviews 16:30HiFi Buys 57:57ecoustics 1:01:20. He does not say a -301 dB signal is audible; he says errors at that level are Head-Fi, Apr 2022ecoustics 49:13.
The distinction itself
In a September 2018 reply on the Watts Up thread, answering a poster who had written about human hearing, he said: "I talk about perception. They are very different things." He defined hearing as the abilities of the ear, "the ear as a transducer is a poor device", and perception as "the brain processing the information from the ears to create the audio illusion that we enjoy" Head-Fi, Sep 2018.
The same split appears earlier and later. In his first Watts Up blog post (March 2016) he wrote: "Hearing is not about how well our ear's work, but is much more about the brain processing the data." Head-Fi, Mar 2016. In August 2016 he wrote that "nobody understands how the brain does the processing it does to separate sounds out into discrete entities and create placement data too" Head-Fi, Aug 2016. In October 2017 he wrote that "The ear is actually a very crude transducer" whose limits the brain's processing "seemingly overcomes" Head-Fi, Oct 2017. In seminars and interviews he says that what we hear is brain construction, not ear output: "What we hear Is not the output from the ears, but the brain processing and interpreting what we're hearing." (October 2017) miceblue425 1:28; "What we hear is not our ears." (CanJam NYC 2022 seminar, video published March 2022) MFG Audio Reviews 15:43; "it's not about the ear, it's about the brain, and what the brain is doing to the data that the ear is giving you" (podcast published October 2026, recording date not stated) Part-Time Audiophile 18:37.
He describes the brain's tasks as separating sounds into individual entities, placing them in space, and judging timbre (a slide talk at Chord's Pump House, published by Hifistatement in October 2017, recording date not stated; and May 2024), and he names transients as a cue for these and for the starting and stopping of notes (March 2016) Hifistatement 43ConvinceMeAudio 4:54Head-Fi, Mar 2016.
Why he distrusts psychoacoustic thresholds
2014 to 2015 (Hugo and Dave). In October 2014, in a post marking a year since what he called the first listening test of Hugo's finished code (a 2017 post dates his first listening to Hugo to October 2012 Head-Fi, Jun 2017), he wrote that he had learned "extremely small errors (errors way below the threshold of audibility) are capable of upsetting the brain's processing of sound and thus become very significant", and that you cannot make assumptions about which errors are audible, so he now does more listening tests in a new design Head-Fi, Oct 2014. In November 2015, writing about Dave, he said that his earlier target for timing accuracy (under 1 uS) was based on the interaural delay figure, not on how the brain processes ear data, and that "science has no idea how the brain separates instruments out" Head-Fi, Nov 2015.
2016 to 2018. In the March 2016 blog post he spelled out the argument. Designing to psychoacoustic thresholds rests on studies using sine waves; the assumption behind it, he wrote, is that "the brain is working at the same resolution as our ears", while science has no understanding of how the brain decodes the ear's data. He compared the ear to CD resolution, noted that an FFT can reveal a -144 dB signal in 16-bit data, and said "the brain probably uses correlation routines to separate sounds out". He offered this as "my explanation for why I have often reliably heard errors that are well below the threshold of hearing", and in the same post called listening tests "very hard and extremely easy to get wrong" Head-Fi, Mar 2016.
At CanJam RMAF in October 2017 he put it in design terms. The normal assumption, he said, is that anything smaller than the perceptual thresholds can be ignored, but he had "often heard errors that are well below the threshold of audibility", citing errors at minus 140 and minus 150 dB (a figure; check the recording) miceblue425 3:50. He said the ear cannot detect such an error on its own, but the brain's processing is interfered with, so no assumptions can be made about how hearing works miceblue425 4:35. In his DAC design masterclass at the same show (given in October 2017, video published December 2017) he said psychoacoustic thresholds "has no bearing on the brain processing" (his grammar) and that "the only way you can do in terms of understanding whether something is audible is to do a very careful, structured listening test" RMAF 54RMAF 1:41.
In January 2018 he gave a Dave example: harmonics at -130 dB were "well below the threshold of audibility", yet when he cut them below -150 dB the difference was "easily audible" Head-Fi, Jan 2018. In the September 2018 reply he said "Psycho-acoustics actually is very simple, involving simple single tones - and we can't base our design requirements on such simple tests", and that psychoacoustic testing has nothing to say on depth, instrument separation or timbre range: "no research has been done on this at all" Head-Fi, Sep 2018. Earlier, on depth in particular, he wrote: "I am not aware of any published psycho-acoustic papers on the technical requirements for depth perception" (June 2017, Davina and noise-shaper context) Head-Fi, Jun 2017. In the 2018 reply he again offered correlation as a mechanism, saying "it is thought that correlation routines are used in perception" and that errors become audible when they "interfere with the brains correlation process" Head-Fi, Sep 2018.
2019 to 2020. In July 2019 he wrote that "The sound-science faction have misunderstood the problem - it's not that the ear/brain can perceive differences that are below the threshold of hearing but it's the interpretation of those differences that's the problem" Head-Fi, Jul 2019. In part 2 of the Passion for Sound interview (published in parts from December 2019 to January 2020; recording date not stated), speaking about depth perception, he said: "if we don't understand how the brain does it, then you can't make assertions as to whether something's going to be audible or not audible" Passion for Sound 15:05. In a podcast published June 2020 (recording date not stated) he called it "dangerous to assume that because the ear is only capable of resolving say minus 90 dB" distortion below that level need not be reduced Part-Time Audiophile 35:33, and said that sine-wave psychoacoustic tests give "quite a crude number in terms of what you need to work towards" and that "it's not about the ear, it's about the brain processing that information" Part-Time Audiophile 36:29. As an example he described a noise shaper with distortion at 200 to 400 kHz; after the fix "you could hear an immediate difference", and "you can't hear 200 kHz, but if a 200 kHz component changes the timing of a transient, then you can perceive it" Part-Time Audiophile 37:19Part-Time Audiophile 38:02. In July 2020 he weighed the correlation idea against a problem: depth is perceived within a few seconds, which limits how much data the brain has to correlate, though he was "sure that correlation routines are an important part of perception" Head-Fi, Jul 2020.
2021 to 2022. In November 2021 he wrote that "the brain compensates for the ears poor technical performance - so the brain can deal with simple distortions. But digital errors are not simple" Head-Fi, Nov 2021. In April 2022, on the Dave thread, he agreed with a poster that needing accuracy at -301 dB "makes no objective sense", then said the brain "does have mechanisms to deal with simple distortion; but it can't deal with small signal distortions, nor noise floor modulation, anharmonic distortions or transient timing distortions", that he follows "the evidence, not what my brain tells me as being sensible or not", and that dismissing his listening evidence without doing one's own tests "is not objective, rational or scientific", while adding "you are right to be sceptical" Head-Fi, Apr 2022. The next day he wrote that "Psychoacoustics seem stuck on determining the performance of the ear, and not on the problems of the brain creating the audio illusion", that he had not seen noise floor modulation in the psychoacoustic literature, and "How the distortion becomes audible I do not understand" Head-Fi, Apr 2022. In an interview published in March 2022 he said: "you need to do very careful listening tests and not assume that something very tiny is inaudible" article, Mar 2022. In September 2022 he wrote that people who say errors well below the ear's threshold are inaudible are "either too lazy to bother to listen objectively, or lack acuity to hear these effects" Head-Fi, Sep 2022.
2023 to 2026. At CanJam NYC 2023 (video published March 2023) he described a common design rule (exceed the ear's performance and the job is done) and said psychoacoustic tests of the ear "are only looking at the ear", with "They're not looking at how the brain is functioning" The Headphone Show 37:31The Headphone Show 37:51. In September 2023 (CanJam SoCal) he called his -301 dB test signal "completely inaudible", adding that "the quality of that signal, if it's slightly wrong, the brain can perceive it" The Source AV 8:39. In March 2024 he asked how such small errors can be perceivable and answered: "Honest answer: I don't know." His guess was that the brain uses autocorrelation to separate instruments MFG Audio Reviews 16:30. In July 2024 he retold the story of engineers who argued that anything below the threshold of audibility could be ignored; he said "even though the error is smaller than the threshold of audibility, the brain perceives the error" HIFICLUB 22:44HIFICLUB 23:24. In a podcast recorded in October 2025 and published in March 2026 he said "a minus 301 dB is never going to be audible, but a minus 301 dB error is perceivable" and called the ear "a crappy transducer" ecoustics 49:13ecoustics 49:59. In the same podcast he said the brain is "Probably" using some kind of correlation routine to calculate depth, and that "in one sense, I don't care because all I want is better sound" ecoustics 43:20; later, on noise floor modulation, he said "this is definitely a case of correlation" ecoustics 1:01:20. In an interview published October 2026 (recording date not stated; its transcript places the recording before May 2026) he said "I suspect that the brain is using correlation techniques to separate individual sounds out", and added that "the ear is acting as a dithered source" so very small errors come through HiFi Buys 57:57HiFi Buys 58:47.
The -180 dB and -200 dB noise shaper story
He tells one anecdote repeatedly: in March 2016, October 2017, September 2018 and July 2024 Head-Fi, Mar 2016Head-Fi, Oct 2017Head-Fi, Sep 2018HIFICLUB 23:24. After arguing with engineers at a semiconductor client company over whether sub-threshold errors mattered, he had one of their engineers, not an audiophile, listen in his home to a noise shaper at -180 dB against one at -200 dB Head-Fi, Sep 2018HIFICLUB 22:44HIFICLUB 23:24. In the September 2018 version Watts said "I conducted the tests single blind", that both offered perfect performance "from a measurement or hearing psycho-acoustic POV", and that the engineer "could readily hear the difference" Head-Fi, Sep 2018. The 2016, 2017 and 2024 versions read here do not mention blinding; each says the engineer was struck by how easy the difference was to hear Head-Fi, Mar 2016Head-Fi, Oct 2017HIFICLUB 24:09. These are his own accounts; the items give no test data.
The interaural delay figure: hearing resolution versus perception
A related distinction concerns timing. In June 2014 he explained that the interaural delay mechanism has a resolving power "of the order of 4uS" and "is only used for left right localisation of sounds" Head-Fi, Jun 2014. In July 2014 he said he used the 4 uS number as an illustration of the importance of timing and had "no idea what levels of timing errors are significant" Head-Fi, Jul 2014. In November 2015, on Dave, he wrote that there is no real limit to "how accurate the time domain needs to be", and that designers should target nS rather than uS Head-Fi, Nov 2015. In March 2018 he said that although 4 uS is the limit of the interaural delay, "the ear/brain is actually much more sensitive to timing errors than just the 4uS", since otherwise the 16FS and 256FS WTA filters would not sound different Head-Fi, Mar 2018. In January 2019 he wrote that 20 years earlier he would have given 4 uS as the requirement, but "hearing resolution and perception are not the same things", 4 uS tells us nothing about pitch, timbre or instrument separation, and "I now know for certain that 4uS resolution is not good enough" Head-Fi, Jan 2019. In July 2019 he wrote that the ear/brain is "at least as sensitive to 4 uS" (adding in parentheses that his subjective evidence suggests it is "actually a great deal more sensitive") and that, as a limit he was researching, "there is no point in reconstructing transients to a higher accuracy than the ear/brain can detect" Head-Fi, Jul 2019. In part 4 of the Passion for Sound interview (published January 2020; recording date not stated) he said the 4 microsecond figure is not the same as timbre, pitch, bass and instrument separation Passion for Sound 2:05.
From 2023 he adds a claim about recent research on the interaural delay itself. In January 2023 he read a paper as suggesting that "the smallest difference in timing is resolvable by the brain" Head-Fi, Jan 2023. In a February 2023 talk he said "Recent research has shown that actually 4 microseconds worth of delay is not the accurate number, but actually there's no resolution there" Dusty Audio 36:57. In March 2023 he said recent research shows "there's no limit" Eric's Hi-Fi 6:09; in June 2024, on the Quartet M Scaler thread, that "there is no resolution limit" and that this agreed with his own listening tests on WTA filters Head-Fi, Jun 2024; in May 2025, and again in the HiFi Buys interview published October 2026 (recording date not stated), he repeated it, citing recent research papers Audiophile Junkie 2:39Audiophile Junkie 3:24HiFi Buys 17:47. In January 2024 he also wrote that he had scoped out how to minimize transient timing errors "to levels where the errors become inaudible or insignificant" Head-Fi, Jan 2024. This archive does not name the research, except that the January 2023 post quotes a passage from a paper without giving its title.
How he says he guards against fooling himself
He does not present skeptics' worries as groundless. In the March 2016 blog he listed placebo, expectation, fatigue, price and brand bias, advised blind tests and said "My rule of thumb is if I can measure it, and it is signal dependent error, then its audible" Head-Fi, Mar 2016. In June 2017, on the charge of confirmation bias in his non-blind depth tests, he wrote "I can't rule that out as a possibility", and stated a plan to publish two recordings with different noise-shaper settings under random file names for a blind depth test Head-Fi, Jun 2017; this archive does not report the outcome. In September 2018 he agreed that sighted-test biases "must be removed or taken into account", said some of his tests show no change, and said "when something extraordinary comes along I do perform single blind tests too"; his listening tests "are about proving to me that something is audible and important", not about convincing others Head-Fi, Sep 2018. He also reports listening tests that found no difference: in February 2025 he wrote that a -1 dB or so drop at 20k with its phase shifts proved "inaudible (at least to my old ears)" Head-Fi, Feb 2025.
He also cites psychoacoustic findings himself, on transients. He says the importance of transients to perception "has been in the text books for many decades" Head-Fi, Aug 2022, cites psychoacoustic tests where removing transients made a piano and a trumpet hard to tell apart Passion for Sound 5:28, and wrote in July 2019 that "We also know from psychoacoustics that transients are essential from a perception POV" Head-Fi, Jul 2019. The objection he states is to basing design requirements on simple threshold tests of the ear Head-Fi, Sep 2018The Headphone Show 37:51.
What he has not said
Within these items he does not say a -301 dB signal, or any error below the ear's threshold, can be heard on its own; he says its effects can be perceived miceblue425 4:35Head-Fi, Apr 2022The Source AV 8:39ecoustics 49:13. On the mechanism he has said "How the distortion becomes audible I do not understand" (April 2022) and "Honest answer: I don't know" (March 2024), and he usually frames correlation as probable, a guess or a suspicion Head-Fi, Apr 2022MFG Audio Reviews 16:30HiFi Buys 57:57. He is firmer at times: in July 2020 he was "sure" correlation routines are an important part of perception while listing a difficulty with the idea, and in the October 2025 podcast he called noise floor modulation "definitely a case of correlation" Head-Fi, Jul 2020ecoustics 1:01:20. The items contain no independent test data on his sub-threshold claims.
Quotes
“you talk about human hearing, but I don't actually talk about human hearing - I talk about perception. They are very different things.”2018-09-27 Head-Fi, Sep 2018
“extremely small errors (errors way below the threshold of audibility) are capable of upsetting the brain's processing of sound and thus become very significant”2014-10-20 Head-Fi, Oct 2014
“Psychoacoustic thresholds, which are the normal things that people work towards in terms of designing products, has no bearing on the brain processing.”2017-10 RMAF 54
“it's not about the ear, it's about the brain processing that information.”2020-06-25 Part-Time Audiophile 36:29
Dated positions
| When | Products | Position |
|---|---|---|
| 2014-06 to 2014-07 | Hugo (original) | Says the interaural delay resolves about 4 uS but is used only for left-right localization, and that he does not know what levels of timing error are significant. Head-Fi, Jun 2014Head-Fi, Jul 2014 |
| 2014-10 to 2015-11 | Hugo (original), DAVE | Says extremely small errors below the threshold of audibility can upset the brain's processing, so audibility cannot be assumed; says his earlier timing target was based on the interaural delay rather than on how the brain processes ear data, and that there is no real limit to the time-domain accuracy needed. Head-Fi, Oct 2014Head-Fi, Nov 2015 |
| 2016-03 to 2016-08 | general | Argues that threshold-based design assumes the brain works at the ear's resolution, which science cannot support, and offers correlation routines ("probably") as his explanation for hearing errors below the threshold. Head-Fi, Mar 2016Head-Fi, Aug 2016 |
| 2017-06 to 2018-01 | general, DAVE | Says psychoacoustic thresholds have no bearing on brain processing, that he has heard errors around minus 140 to minus 150 dB and that cutting -130 dB harmonics on Dave was easily audible, and that only structured listening tests can show audibility. Head-Fi, Jun 2017miceblue425 3:50miceblue425 4:35RMAF 54RMAF 1:41Head-Fi, Jan 2018 |
| 2018-03 to 2019-01 | general, Hugo M Scaler | Says he talks about perception, not hearing; that psychoacoustic testing uses simple tones and says nothing on depth, separation or timbre range; and that 4 uS interaural resolution is not good enough for timing in perception. Head-Fi, Sep 2018Head-Fi, Mar 2018Head-Fi, Jan 2019 |
| 2019-07 to 2020-07 | general, Hugo M Scaler | Says the real problem with listening is interpreting sub-threshold differences; that without understanding the brain one cannot assert what is audible; calls sine-wave threshold results a crude design target; and says he is sure correlation matters while noting a difficulty with the idea. Head-Fi, Jul 2019Head-Fi, Jul 2019Passion for Sound 15:05Part-Time Audiophile 36:29Head-Fi, Jul 2020 |
| 2021-11 to 2022-09 | general, DAVE | Says the brain copes with simple distortion but not small-signal, noise floor modulation, anharmonic or transient timing errors, that psychoacoustics focuses on the ear, that he does not understand how the distortion becomes audible, and that careful listening tests are needed rather than assuming tiny errors are inaudible. Head-Fi, Nov 2021Head-Fi, Apr 2022Head-Fi, Apr 2022article, Mar 2022Head-Fi, Sep 2022 |
| 2023-01 to 2026-10 | general, Quartet M Scaler | Cites recent research for there being no resolution limit to the brain's use of interaural timing; says a -301 dB signal is inaudible but errors at that level are perceivable; frames correlation mostly as a guess or suspicion, though once in a podcast recorded October 2025 as "definitely" the case for noise floor modulation. Head-Fi, Jan 2023Dusty Audio 36:57The Source AV 8:39MFG Audio Reviews 16:30Head-Fi, Jun 2024ecoustics 49:13ecoustics 1:01:20HiFi Buys 57:57 |
Caveats and open points
- Figures taken from them (minus 140 and minus 150 dB, 200 to 400 kHz, 4 microseconds, minus 180 and minus 200 dB, minus 301 dB) should be checked against the recordings.
- Everything here is Watts's own account. The items contain no independent measurement or controlled-test data supporting or refuting his claim that errors below the ear's threshold are perceivable, and his anecdotes (the noise shaper test with an engineer) are reports, not data. Only the September 2018 telling of that anecdote says the test was single blind.
- His framing of the correlation mechanism varies: mostly hedged ("probably", "it is thought", "my guess", "I suspect"), but "sure" it is an important part of perception in July 2020 and "definitely a case of correlation" for noise floor modulation in a podcast recorded October 2025 (published March 2026). He has also said he does not know how sub-threshold errors become audible (2022, 2024).
- His statements that recent research shows no resolution limit for interaural timing are cited without naming the research, except for a January 2023 post that quotes a passage from a paper without its title; readers should not treat the research as identified.
- On timing he describes his own view moving: in 2015 and 2016 he wrote that he had underestimated the brain's sensitivity and that his old 1 uS target was not enough, and in January 2019 that 20 years earlier he would have said 4 uS. In July 2019 and January 2024 he still wrote of a level beyond which transient timing accuracy stops mattering, alongside 2022 and 2024 statements that any program-dependent timing error, however small, seems perceivable.
- The German Hifistatement interview text is the publisher's translation of English answers and is not quoted; the one Hifistatement item cited is a transcribed English audio clip. Quotes keep his British spelling and punctuation, including original typos.
- Where a recording date is not stated, talks and interviews are dated here by publication, and the text says so; the order of such items is the order of publication, not necessarily of speaking.
Topics: Auditory processing: how the brain hears, Measurement vs listening