[pulseaudio-discuss] Resampler quality evaluation results

[Alexander E. Patrakov]

25.08.2014 00:53, Alexander E. Patrakov wrote:
> Unfortunately, there is a bug on win81 plots, because Windows Media
> Player by default attenuates the file by 6 dB, and my scripts compensate
> for that, but also amplify the quantization noise. I am too lazy to fix
> this today. Please shift the whole win81-wmp curve down by 6 dB, and
> you'll hopefully get an approximately correct result.

Yesterday's discussion with David Henningson on IRC has led to the 
discovery of another bug. Namely, all plots are saturated at 18 dB. That 
was because I mistakenly included the distortions introduced by the 
resampler into both the signal and the distortion. So the original plot 
answers the question: how much audible are the distortions similar to 
the original ones, in the presence of both the original sine wave and 
the original distortions? As the distortions _also_ mask a copy of 
themselves, the answer cannot be greater than 18 dB.

This is now fixed in git, but this also means that I can no longer plot 
the trivial resampler using the same convenient y scale as others. At 
very high frequencies (20 kHz), the distortion audibility of that 
resampler goes up to 80 dB, and that's completely understandable. At 
such high frequencies, the distortions of this resampler are indeed more 
powerful than the original signal, and, unlike the original signal, are 
concentrated in the frequency range that we hear very well, unlike the 
original signal (which no longer masks anything).

Results for high- and medium-quality resamplers (i.e. anything better 
than speex-float-3 below 10 kHz) are not significantly affected by the bug.

-- 
Alexander E. Patrakov