What is a transient timing error, and why is it audible when a fixed delay or phase shift is not?

What he means by a transient timing error

The core definition recurs from 2014 to 2026. Sampling theory says an infinite-length sinc interpolation filter reconstructs a bandwidth-limited signal perfectly, including a transient that starts between samples; once the tap count is cut, "the precision of the timing of the transient in the middle falls off" (Hugo thread, 2014-04-17) Head-Fi, Apr 2014. His fullest worked example (2016-05-17): a signal sampled at 44.1 kHz crosses zero 20,155 nS after a sample; an ideal sinc filter reproduces the crossing at exactly 20,155 nS, but a finite, non-sinc filter might put it at 19,000 or 21,000 nS, and "it is these differences in the timing of transients, are what I am talking about" Head-Fi, May 2016. He repeated the example in August 2016 with a 15,679 nS crossing and said the error lies in the interpolation filter, as distinct from input jitter, which his DPLL removes Head-Fi, Aug 2016.

He repeatedly stresses that the error is not a constant. In 2017 he described it as "the non-linear timing uncertainty that the interpolation filter introduces" and said that when people talk about timing errors they think of linear things like ringing, which he is not interested in Head-Fi, Feb 2017; in 2016 he wrote that by timing accuracy he does not mean ringing performance Head-Fi, May 2016. In 2018: "Its the dependence on sampling that creates the sound quality problems" Head-Fi, Feb 2017, and the error is a matter of "the timing of transients is non-linear, and constantly changing back and forth dependent upon the sampling point and the past and future samples" Head-Fi, Aug 2018. He defines a transient broadly: "I define a transient as everything that is not steady state" (2018-09-24) Head-Fi, Sep 2018.

In talks in 2018, 2022 and 2024 he says the signal before and after a transient modulates its timing Pursuit Perfect System 16:26MFG Audio Reviews 17:42ConvinceMeAudio 6:51; in 2022 he put it as instruments modulating the timing of other instruments MFG Audio Reviews 17:42. In November 2021 he wrote that a correctly built filter has timing errors independent of amplitude, but that in a real filter individual instruments' transients modulate the timing of other instruments' transients, and only sinc coefficients avoid this Head-Fi, Nov 2021. The most recent statements read for this answer say the same thing: transients "modulate backwards and forwards depending upon what's happening in the future, what's happened in the past and the filter structure itself" (podcast published 2026-10-14, recording date not stated) Part-Time Audiophile 28:11. In an interview about the Quartet published October 2026 he said his research found two different effects degrading transient timing in all interpolation filters, and "they’re contrary to one another" article, Oct 2026.

He also names a second, amplitude-related kind. In 2016 he said timing problems arise when timing varies with amplitude (a small transient timed differently from a large one) or with when the ADC sampled it Head-Fi, Oct 2016. He attributes the amplitude kind to DSD noise shapers, where a small transient takes longer to get through the quantizer (several microseconds for DSD64 in his simulations, 2015-09-25 Head-Fi, Sep 2015; tens of microseconds, 2017-03-15 Head-Fi, Mar 2017), and to PWM, where higher amplitude outputs come earlier Head-Fi, May 2016Head-Fi, Jun 2022. In 2018 he called the filter examples sampling-related timing uncertainty and DSD an amplitude-related one Head-Fi, Jul 2018.

How big he says the errors are

His illustration is a 20 kHz sine burst sampled at its start. In 2014 he said a one-tap filter gives a transient timing error of hundreds of microseconds Head-Fi, Jul 2014; in 2017 he put it at 100uS or so Head-Fi, Jan 2017. In talks he gives roughly 100 microseconds for minimum phase and IIR filters (CanJam RMAF 2017) miceblue425 12:06 and for NOS, IIR, minimum phase and MQA filters (Malaysia 2025) Audiophile Junkie 5:21; those two figures come and should be checked against the recordings. In 2015 he wrote that conventional short tap length filters give timing errors of tens of microseconds Head-Fi, Mar 2015. When a poster objected that the burst's sharp start is not a bandwidth-limited signal, he agreed, said he chose it to get the idea across to Mojo users, and wrote: "The illustration is just to show how timing uncertainty comes about." (2017-02-13) Head-Fi, Feb 2017.

In 2017 and 2021 he described the error by its area, which falls by a factor of four each time the sample rate doubles Head-Fi, Dec 2017Head-Fi, Sep 2021. In 2021: "It seems that the ear/brain is sensitive to the area of the transient error" Head-Fi, Sep 2021. He said this is why higher sample rates sound better, and that the reconstruction problem applies equally to recordings with and without ADC aliasing Head-Fi, Sep 2021.

Why a fixed delay or phase shift is not audible, in his account

His explicit statements on this side run from 2016 to 2026. In October 2016 he said phase errors and timing differences with frequency response are linear: "these are linear errors, and are inconsequential - the brain is used to dealing with these kinds of problems" Head-Fi, Oct 2016 (the scraped text drops a word just before this passage). In July 2017 he said he is not worried about phase non-linearity in the traditional sense: "I have no evidence that this type of behavior is important subjectively (unless it's gross in which case FR would be affected too)", whereas with the timing he means, "the key is that the timing is constantly changing" Head-Fi, Jul 2017.

His plainest examples come from answering posters. In August 2017, on a 40 kHz delay of about 1cm: "a linear (unvarying) shift of 1cm would be hard (impossible) to hear", but if the delay is constantly and randomly changing, "then you absolutely would hear the change, as the brain uses timing information to perceive pretty much everything" (Hugo 2 thread) Head-Fi, Aug 2017. In June 2019: "A fixed timing error is not important - so a shift of 1us at 5 kHz say is inaudible; but a 1us shift that is constantly changing is audible" Head-Fi, Jun 2019. In March 2019 he noted that "all symmetric FIR filters are phase linear, that is they have a constant delay irrespective of frequency", and that the WTA filters are about the timing of transients rather than phase shifts as such Head-Fi, Mar 2019.

He hedges on conventional phase response. In July 2020 he said he has no evidence it matters so long as it is the same on both channels, but "I am open minded or agnostic on it", then added that a phase-linear HF filter and an older one with a phase shift and -0.8dB at 20k: "They sounded pretty much the same" Head-Fi, Jul 2020. In the same post: "What is an enormous issue is a phase shift that varies with time", or with amplitude Head-Fi, Jul 2020. In January 2021 he said he has no evidence that a simple, amplitude-linear phase shift is important unless frequency response also changes strongly Head-Fi, Jan 2021. In July 2021: "it's not the absolute value of timing that is important, but that the timing does not vary with amplitude (giving small signal errors) or with sampling (giving transient timing errors)" Head-Fi, Jul 2021. He links this to the brain subtracting the acoustic footprint of a room Head-Fi, Jul 2020Head-Fi, Jul 2021.

Before his later tests, in April 2019, he guessed that an FIR decimation filter for his ADC project would sound closer to the original than an IIR one, because the IIR filter changes the in-band harmonics Head-Fi, Apr 2019. He presents two later listening tests as proof. In June 2024 he said the Mojo 2 HF filter, a 50kHz 4th order IIR, changes the timing of transients linearly, but "linear changes don't matter, as the delay is unvarying with amplitude or time"; with his FIR ADC bandwidth-limiting filter in place, switching the Mojo 2 filter in made no audible change Head-Fi, Jun 2024. In February 2025 he described a similar test of an IIR bandwidth-limiting filter, concluding that "a -1dB or so at 20k with attendant phase shifts are inaudible (at least to my old ears)" Head-Fi, Feb 2025. In an interview published October 2026 (its transcript places the recording before May 2026; recording date not stated): "A linear phase shift, in my experience, is inaudible." HiFi Buys 1:58:09 He went on to say that a timing that is rock solid, even if wrong in a linear or consistent fashion, won't matter; what matters is that it does not vary HiFi Buys 1:58:18.

He applies the same split to analog parts. Inductors and ferrites are non-linear, so their phase shift changes with signal level and the timing of transients becomes amplitude dependent, which he says is audible (2017, 2018, 2021) Head-Fi, Jul 2017Head-Fi, Feb 2018Head-Fi, Apr 2021. The phase-with-amplitude work that came after the M Scaler concerns depth; he says the WTA filters pass that steady-state test Head-Fi, Apr 2022Head-Fi, Mar 2023.

Why he says a varying error is audible

His explanation is perceptual. He says the brain uses transients for left-right placement, bass pitch and timbre, and that errors in their timing degrade each (his CanJam presentation, posted 2018-07-26) Head-Fi, Jul 2018. Because the error depends on when the signal was sampled and on past and future signals, "it's this constant change in the timing of transients that confuses the brain" Head-Fi, Jul 2018. He describes the result as softness: "when you get these huge transient errors, the brain can't make sense of the audio", it fails to perceive the transients, and things sound softer, like an image going out of focus Head-Fi, Jul 2018; he said the same of NOS filters in 2016 and 2024 Head-Fi, Oct 2016HIFICLUB 29:50.

He compares the error to jitter. In 2014 he wrote "timing error is the same as jitter" Head-Fi, Jul 2014; in 2017, "The transient timing error from an interpolation filter can be thought of exactly like jitter", and "The instance that nonlinearities come into play means that you can't relate time domain to the frequency domain." Head-Fi, Oct 2017. In 2016 he said the brain copes with harmonic distortion and noise because the ear itself is non-linear, but "it has no capability to deal with timing errors from sampled data as this is a new problem" Head-Fi, Aug 2016; in 2015 he wrote that digital aberrations upset the brain, which is not used to dealing with timing errors Head-Fi, Dec 2015.

How small an error he says is audible

Here he says himself that his view changed. In July 2014 he wrote: "I have no idea what levels of timing errors are significant" Head-Fi, Jul 2014. In May 2016 he wrote that he once thought accuracy to a microsecond was perhaps OK, but "now I know that instead of worrying about uS we need to worry about getting it correct to nS's", citing the audible change in Dave from moving a 16FS WTA stage to 256FS Head-Fi, May 2016. In September 2015 he had already written that the brain is much more sensitive than the order of 4 uS, the inter-aural delay figure Head-Fi, Sep 2015. In October 2016 he put the requirement at tens of nS Head-Fi, Oct 2016. In January 2017: "I used to think 1uS errors would be inaudible, now I take the view that any timing error (where a transient is slightly earlier or later against the rest of the signal) of any size, no matter how small, is important." Head-Fi, Jan 2017. At CanJam RMAF in October 2017 he said he had thought 4 microseconds was the target, "I actually was wrong about that", and gave the order of 100 nanoseconds or less miceblue425 10:26.

His posts from 2018 to 2026 keep the any-size position. In September 2018: "my current assumption is that any transient timing error that is constantly varying is likely to be audible, no matter how small that error is", based on hearing half a million to one million WTA taps and the 256FS stage Head-Fi, Sep 2018. In an interview published January 2020 (recorded by December 2019; recording date not stated) he said he could not yet put a number on it, suspecting even 10 nanoseconds of uncertainty would be audible Passion for Sound 5:03. In August 2022: "If it was just a uS problem, you would not be able to hear much of a difference from 100k taps WTA to 1M taps." Head-Fi, Aug 2022. In July 2022 he called "the level of transient timing reconstruction accuracy that is needed" the only contentious aspect Head-Fi, Jul 2022. From 2023 he links this to research on the interaural delay; in June 2024: "recent research indicates that the brain can perceive any change in timing - there is no resolution limit" Head-Fi, Jan 2023Head-Fi, Jun 2024. In a podcast recorded in October 2025 and published in March 2026, on his earlier 16-bit reconstruction target: "If there's the tiniest, tiniest error in terms of timing reconstruction, then it has audible consequences." ecoustics 4:49.

What he has not said: he gives no measured threshold at which a varying timing error becomes inaudible. He said in 2017 and 2018 that he had no numbers for the timing error itself, and in an interview published January 2020 (recording date not stated) that he could not yet put a number on the accuracy needed Head-Fi, Feb 2017Head-Fi, Sep 2018Passion for Sound 5:03. He gives no number for how large a fixed delay or linear phase shift can be before it matters beyond his examples (1cm at 40 kHz, 1us at 5 kHz, -1dB at 20k with its phase shift), and he calls himself agnostic on conventional phase response Head-Fi, Aug 2017Head-Fi, Jun 2019Head-Fi, Feb 2025Head-Fi, Jul 2020.

Quotes

“the timing of transients is non-linear, and constantly changing back and forth dependent upon the sampling point and the past and future samples”2018-08-04 Head-Fi, Aug 2018

“A fixed timing error is not important - so a shift of 1us at 5 kHz say is inaudible; but a 1us shift that is constantly changing is audible”2019-06-14 Head-Fi, Jun 2019

“But it's a linear change, and linear changes don't matter, as the delay is unvarying with amplitude or time.”2024-06-11 Head-Fi, Jun 2024

“my current assumption is that any transient timing error that is constantly varying is likely to be audible, no matter how small that error is.”2018-09-24 Head-Fi, Sep 2018

Dated positions

Caveats and open points

• His statements that fixed delays and linear phase shifts are inaudible rest on his own listening tests and on the absence of evidence to the contrary; he describes himself as agnostic on conventional phase response (2020).
• His 20 kHz tone-burst illustration uses a signal that is not bandwidth limited; he acknowledged this in 2017 and called it a simple illustration.
• Head-Fi, Oct 2016 has a word dropped by the scraper just before his statement that phase and frequency-response timing errors are linear and inconsequential; the quoted passage after the gap is intact.
• The Weekly ASCII (2016) and Hifistatement (2024) articles cover the same ground but are a Japanese paraphrase and a German publisher's translation; they are not quoted here.
• He says the brain needs time to adjust to timing improvements (a hypothesis he labels as such, 2016); that claim is not covered here.

Topics: Transient timing error from reconstruction, Error classes and which are audible