Let's develop an ASR inter-sample test procedure for DACs!

[Scytales]

Here is something that may be of interest to all of you who wonder if inter-sample overs are really a thing in actual music.

In August 2025, the famous blogger @Archimago published a survey he has conducted over his own digital library of about 107.000 2.0 stereo music tracks and over 7.000 multichannel tracks to search in each and every one of them the highest inter-sample peaks calculated with the foobar True Peak Scanner add-on.

Here are the results of his work : click here the read the article on Archimago's blog.

 

[NTK]

Here is something that may be of interest to all of you who wonder if inter-sample overs are really a thing in actual music.

In August 2025, the famous blogger @Archimago published a survey he has conducted over his own digital library of about 107.000 2.0 stereo music tracks and over 7.000 multichannel tracks to search in each and every one of them the highest inter-sample peaks calculated with the foobar True Peak Scanner add-on.

Here are the results of his work : click here the read the article on Archimago's blog.

I completely concur with his conclusion.

 

[NTTY]

Very well documented, as usual!
Well, so here we go, at least on a technical perspective, audibility left aside. It is interesting to know how a DAC/Player will handle these cases, and what to do to prevent it, if possible.

 

[VientoB]

Has anyone listened to that track with 7.72dbTP?

 

[tmtomh]

I completely concur with his conclusion.
View attachment 477334

I agree this is the best practice. And if the production of music had the same level of engineering involvement and ability to standardize and enforce specifications as the production of audio hardware does, it's likely there would be some kind of engineering society standard that would guide creating digital masters that did not produce intersample overs (and that such standard would be widely respected).

However, I don't see that as the case. And so I would say this is a situation more akin to how engineers design audio components to try to withstand end-user abuse - relays, protection circuits, tape auto-stop mechanisms, cassette recording-prevent tabs, and so on.

So in my view, knowing what we know, it should be considered a basic tenet of good DAC design to make them tolerant of a reasonable level of intersample over. +6DB would probably cover pretty much every case, but I would say a reasonable compromise standard would be +3dB.

With all that said, I do think this might be a non-issue, or at least a conveniently solvable issue, for more and more hi-fi listeners, because it seems to me that more and more of us have access to digital playback systems that include decently designed digital volume controls that can attenuate the signal to prevent analogue IS overs.

In particular, my understanding is that if you use an active speaker system with digital input and a properly designed digital volume control, IS overs are pretty much a non-issue unless you're running the speakers right up to the bleeding edge of their output capability.

 

[VientoB]

I completely concur with his conclusion.
View attachment 477334

I'm not sure I'd say it isnt great sounding hifi music. Depends how audible the intersample overs are.

 

[danadam]

With all that said, I do think this might be a non-issue

For me it always was a non-issue. ReplayGain proposal was published already in 2001:

Replay Gain - A proposed standard

Still, the inability or unwillingness of music production industry to create such standards on their side doesn't reflect good on them.

 

[markstein]

With all that said, I do think this might be a non-issue, or at least a conveniently solvable issue, for more and more hi-fi listeners, because it seems to me that more and more of us have access to digital playback systems that include decently designed digital volume controls that can attenuate the signal to prevent analogue IS overs.

In particular, my understanding is that if you use an active speaker system with digital input and a properly designed digital volume control, IS overs are pretty much a non-issue unless you're running the speakers right up to the bleeding edge of their output capability.

This is true, unless there is the need for SRC which often is done before digital volume control. In this case you have to take care to prevent clipping during DRC (especially upsampling)

 

[tccalvin]

Here is something that may be of interest to all of you who wonder if inter-sample overs are really a thing in actual music.

In August 2025, the famous blogger @Archimago published a survey he has conducted over his own digital library of about 107.000 2.0 stereo music tracks and over 7.000 multichannel tracks to search in each and every one of them the highest inter-sample peaks calculated with the foobar True Peak Scanner add-on.

Here are the results of his work : click here the read the article on Archimago's blog.

I disagree with Archimago's spin on this topic. If we were talking about analog equipment, then sure, who cares if there are peaks at +0.1dB. All that means is that you're going to get a tiny bit more harmonic distortion.

With digital equipment using oversampling, it doesn't matter if the peaks are at +0.1dBFS, +1dBFS, +10dBFS or +100dBFS. It'll all read zero. That means a completely (square-ish) chopped wave at best, and filter overload at worst.

It's ridiculous that we can obsess over DAC SINAD values ranging from 96dB to 120dB when many DACs are constantly doing this (example of filter overload from Benchmark's article):

What's the point of having that pretty 0dBFS 1kHz sine wave graph if you're actually listening to this garbage half the time?

If a recording engineer were to digitally record a signal where every other snare hit is clipping above 0dBFS, it would be unacceptable. They would instantly get fired. If they were to say "who cares. it's only clipping on quick peaks so it's not likely to be audible. it's fine" no one would take them seriously. Why should we have a different standard for audio reproduction devices?

Furthermore, this problem doesn't just affect casual listeners. If you use any digital instrument, effect, outboard gear, or you're doing hybrid mixing utilizing digital and analog gear, all of that distortion is going to be baked in your recording, simply because the signal is going through a DAC. That's unacceptable.

If we do want digital audio to be reproduced as faithfully as possible, we can't ignore inter-sample peaks.

 

[markstein]

It's ridiculous that we can obsess over DAC SINAD values ranging from 96dB to 120dB when many DACs are constantly doing this (example of filter overload from Benchmark's article):

What's the point of having that pretty 0dBFS 1kHz sine wave graph if you're actually listening to this garbage half the time?

If a recording engineer were to digitally record a signal where every other snare hit is clipping above 0dBFS, it would be unacceptable. They would instantly get fired. If he were to say "who cares. it's only clipping on quick peaks so it's not likely to be audible. it's fine" no one would take them seriously. Why should we have a different standard for audio reproduction devices?

If we do want digital audio to be reproduced as faithfully as possible, we can't ignore inter-sample peaks.

+1

 

[Hayabusa]

Sounds unbelievable

I think the proof is, that given arbitrary sequence of sample values, you can get any ISP value you want. Fallacy is that you can't get these values with a proper digitalization of analog signal, because of applying anti-aliasing filter first. Quote: "The pathological waveform we are interested in is a series of

alternations between

and

, followed by silence". Alterations between -1 and 1 is frequency Fs/2, which is not allowed in a proper data, right?

Probably the attempt to digitize single impulse gives maximum value of real world ISP, which would be about 4dB.

It could be interesting to test how does DAC behave with artificial data, this would give some info, how resilient is internal processing. But probably more meaningful would be a test with a proper real signal.

Note that the maximum peak reached is determined fully by the the coefficients and length of the interpolation filter of the tested DAC.

 

[Mnyb]

Here is something that may be of interest to all of you who wonder if inter-sample overs are really a thing in actual music.

In August 2025, the famous blogger @Archimago published a survey he has conducted over his own digital library of about 107.000 2.0 stereo music tracks and over 7.000 multichannel tracks to search in each and every one of them the highest inter-sample peaks calculated with the foobar True Peak Scanner add-on.

Here are the results of his work : click here the read the article on Archimago's blog.

My takeaway from reading the article , as that this problem is not insignificant. And can’t be ignored, it’s there . Not a rare theoretical edge case .

The 3db suggestion is probably solid , it’s very likely a good trade off i rather loose 3dB sinad .

But been using all manner of streamers and computer audio for a long time I can use digital volume and gain settings to fix this .

So the fix does not necessarily needs to be in the DAC .

 

[danadam]

If a recording engineer were to digitally record a signal where every other snare hit is clipping above 0dBFS, it would be unacceptable. They would instantly get fired. If they were to say "who cares. it's only clipping on quick peaks so it's not likely to be audible. it's fine" no one would take them seriously.

And yet no one fires mastering engineers or producers who slam everything to 0 dBFS and who are actually the ones responsible for creating those ISOs.

Why should we have a different standard for audio reproduction devices?

I don't see the standards to be different. Neither side cares.

 

[KSTR]

If a recording engineer were to digitally record a signal where every other snare hit is clipping above 0dBFS, it would be unacceptable. They would instantly get fired. If they were to say "who cares. it's only clipping on quick peaks so it's not likely to be audible. i

What they say is absolutely correct. When you have a snare producing ISOs all the time, it's already been hard-clipped anyway and the perceived difference of those ISO's being faithfully reproduced or not is non-existent. Even the (rare) worst-case of ISO effects, the dreaded wraparound, is inaudible for this kind of signals. Simply clipped ISO not to speak of...

 

[KSTR]

The 3db suggestion is probably solid , it’s very likely a good trade off i rather loose 3dB sinad .

A good compromise is to (soft-)clip or brickwall fast-limit (with look-ahead) the upsampled signal to remove the ISOs and convert back to target sample rate. No wasted headroom and still pretty much inaudible difference compared to faithfully reproduced ISOs.

 

[Mnyb]

What they say is absolutely correct. When you have a snare producing ISOs all the time, it's already been hard-clipped anyway and the perceived difference of those ISO's being faithfully reproduced or not is non-existent. Even the (rare) worst-case of ISO effects, the dreaded wraparound, is inaudible for this kind of signals. Simply clipped ISO not to speak of...

I suppose there a lot of badly produced music that’s ”pre clipped” that we can’t possibly save .

I can imagine that in sampled and very produced music , it can slip in some mistakes.

 

[KSTR]

I suppose there a lot of badly produced music that’s ”pre clipped” that we can’t possibly save .

Yep. The track "Dear Mr. Man" from Prince's 2004 Album "Musicology" is an example, as is the whole album. In technical terms it would go as badly produced... but it still sounds OK to even good because it is a rather sparse arrangement (typical for Prince) thus not suffering that much from being brickwalled to death, enough air between notes.

 

[morillon]

The enormous irony is that these issues are mainly present with music in very narrow dynamic ranges... easier to manage...
it's crazy...

(Analog and its clipping management made much more sense in this case...)
;-)

Now ,We need to see how this is handled by streaming providers...the key point...

 

[Scytales]

Archimago's findings over his own library inspired me an observation.

Obviously, the musical genre the less affected by inter-sample overs is classical music, ie a genre whose recordings are always done with faithfulness to actual sounds that are produced with mechanical instruments or human voices in mind. In other words, sounds that are actually recognizable by listeners accustomed to attend to live performances, to the point that said listeners would be severely bothered by impairments in sound reproduction that would make a sound short of unrecognizable.

Data provided by Archimago shows that, in the case of classical music, avoiding clipping remains a primary concern for almost all sound engineers.

From my perspective, this is an interesting factual observation about what is considered important by people whose constant concern is to be faithful to an original sound at all times in their daily work.

 

[VientoB]

I disagree with Archimago's spin on this topic. If we were talking about analog equipment, then sure, who cares if there are peaks at +0.1dB. All that means is that you're going to get a tiny bit more harmonic distortion.

With digital equipment using oversampling, it doesn't matter if the peaks are at +0.1dBFS, +1dBFS, +10dBFS or +100dBFS. It'll all read zero. That means a completely (square-ish) chopped wave at best, and filter overload at worst.

It's ridiculous that we can obsess over DAC SINAD values ranging from 96dB to 120dB when many DACs are constantly doing this (example of filter overload from Benchmark's article):

View attachment 477433

What's the point of having that pretty 0dBFS 1kHz sine wave graph if you're actually listening to this garbage half the time?

If a recording engineer were to digitally record a signal where every other snare hit is clipping above 0dBFS, it would be unacceptable. They would instantly get fired. If they were to say "who cares. it's only clipping on quick peaks so it's not likely to be audible. it's fine" no one would take them seriously. Why should we have a different standard for audio reproduction devices?

Furthermore, this problem doesn't just affect casual listeners. If you use any digital instrument, effect, outboard gear, or you're doing hybrid mixing utilizing digital and analog gear, all of that distortion is going to be baked in your recording, simply because the signal is going through a DAC. That's unacceptable.

If we do want digital audio to be reproduced as faithfully as possible, we can't ignore inter-sample peaks.

Well hang on because it's the mastering or recording engineers who created these overs in the first place. And yet they're not fired.