AnalogSteph
Major Contributor
I have been playing with the options for "True peak scan" in the Foobar2000 ReplayGain scanner lately, which allows you to use any resampler DSP installed for the oversampling. This made me notice a marked difference in results between "auto 4x oversample" - which I presume uses the filtering suggested in ITU-R BS.1770-5 (or maybe not?) - and e.g. doing 4x upsampling in the SoX resampler DSP with default settings (quality normal, passband 95%, phase response 50% = linear). As the latter is the resampler I routinely use for resampling to either 48 or 192 kHz, this is not exactly an academic exercise.
For example, here's my CD with the highest peaks I could find, Heathers "Kingdom" from 2012 that was a bit behind the times as it clearly was not mastered using an oversampled brickwall limiter which were gaining widespread acceptance at the time (and dare I say it, it always sounded broken to me), using the SoX resampler DSP as outlined above:
And this is the result with Auto 4x:
They cannot even completely agree about which tracks have the highest peak levels in which order, although there's the same 4 tracks to be found in the top 5. In the case of Underground Beneath, there's a whopping 1.67 dB difference between both upsampling algorithms.
For a slightly more common release, here's Ladyhawke's s/t from 2008, arguably the peak loudness war year (I don't have Death Magnetic, X&Y or other notorious examples as I tended to stay well clear of such records at the time). First SoX 4x:
And then Auto 4x:
Again, there is some disagreement, up to 1.68 dB in this case. (And yes, the SoX resampler DSP with its SSE3 optimizations seems to be quicker than the auto algorithm indeed.)
In short, oops. True peak is clearly a function of DAC (or resampler) reconstruction filter, and those may vary. The advice to master to -1 dBTP may not only be completely justified, but even falling short by a bit.
Now what about some modern releases then? Going all the way to 2024, Caroline Rose's Art of Forgetting is not atypical for what I'm seeing these days. Again, first SoX:
...and then Auto 4x:
Disagreements between algorithms have shrunk to 0.24 dB, and worst-case peak is +0.655 dBFS.
I assume that they were going for 0 dBTP, but clearly they slightly missed the mark. The filter used for TP metering must be less steep than either of the resampler options I tried.
Finally a slightly hotter one, fairly fresh off the presses as it only released about a month ago, Desperate Journalist's No Hero - first with SoX again:
...and then Auto 4x:
That's about a 0.66 dB disagreement on the first track, and a worst-case peak of +1.51 dBFS.
For example, here's my CD with the highest peaks I could find, Heathers "Kingdom" from 2012 that was a bit behind the times as it clearly was not mastered using an oversampled brickwall limiter which were gaining widespread acceptance at the time (and dare I say it, it always sounded broken to me), using the SoX resampler DSP as outlined above:
And this is the result with Auto 4x:
They cannot even completely agree about which tracks have the highest peak levels in which order, although there's the same 4 tracks to be found in the top 5. In the case of Underground Beneath, there's a whopping 1.67 dB difference between both upsampling algorithms.
For a slightly more common release, here's Ladyhawke's s/t from 2008, arguably the peak loudness war year (I don't have Death Magnetic, X&Y or other notorious examples as I tended to stay well clear of such records at the time). First SoX 4x:
And then Auto 4x:
Again, there is some disagreement, up to 1.68 dB in this case. (And yes, the SoX resampler DSP with its SSE3 optimizations seems to be quicker than the auto algorithm indeed.)
In short, oops. True peak is clearly a function of DAC (or resampler) reconstruction filter, and those may vary. The advice to master to -1 dBTP may not only be completely justified, but even falling short by a bit.
Now what about some modern releases then? Going all the way to 2024, Caroline Rose's Art of Forgetting is not atypical for what I'm seeing these days. Again, first SoX:
...and then Auto 4x:
Disagreements between algorithms have shrunk to 0.24 dB, and worst-case peak is +0.655 dBFS.
I assume that they were going for 0 dBTP, but clearly they slightly missed the mark. The filter used for TP metering must be less steep than either of the resampler options I tried.
Finally a slightly hotter one, fairly fresh off the presses as it only released about a month ago, Desperate Journalist's No Hero - first with SoX again:
...and then Auto 4x:
That's about a 0.66 dB disagreement on the first track, and a worst-case peak of +1.51 dBFS.
Last edited:
Is that lossless or MP3? (Also, what sort of monstrous CPU do you have there? 5950X? 7950X? Threadripper? 13900K? 14900K? Some Xeon?)
