http://bugs.winehq.org/show_bug.cgi?id=14717
--- Comment #22 from Alexander E. Patrakov patrakov@gmail.com 2010-11-26 10:36:41 CST --- I have built Wine with the second version of your patch. The sound is not horrible (in fact, I cannot distinguish it from the one from the best resampler), but, according to both your plot and the following unscientific test, your resampler is definitely not the best.
Now the test:
1) Install foobar2000 in wine and JAAA in linux
2) Configure Wine to output sound through JACK at either 96 or 192 kHz sampling rate, as permitted by your hardware
3) Run foobar2000, play a CD rip. Route its audio output to "jaaa -j" using something like qjackctl or patchage
4) Inspect the spectrum of wine output in realtime using JAAA. Use the "peak hold" option. You will see that, on typical contemporary music, the ultrasound components after your resampler peak at something like -80 dB. This also agrees with your plot.
Yes, I know that I should really look at distortions below 20 kHz, not at ultrasound components, but there are no such easy-to-use tools for it.
5) Now enable the resampler plugin in foobar2000's DSP manager (so that your resampler is out of the way now) and look at the spectrum again. The ultrasound components are at -120 dB now.
Moreover, without the "peak hold" option, it is obvious that foobar's ultrasound is noise + occasional glitches, while your resampler's ultrasound is visibly correlated with the sound that is piped through it.
The default ALSA resamplers can also be compared with your resampler this way. Use an .asoundrc that routes the audio output from ffplay through a plug plugin and then to jack. All of the working ones (pph aka speex, and libsamplerate) keep the ultrasound noise at -120 dB.