http://bugs.winehq.org/show_bug.cgi?id=14717
--- Comment #18 from Alexander E. Patrakov patrakov@gmail.com 2010-11-24 12:10:19 CST --- As for the transition band width vs impulse width vs stopband attenuation trade-off: I think it is necessary to study how these parameters are chosen by other resamplers out there (i.e., the speex resampler, libsamplerate and libavcodec) so that our resampler doesn't stand out either as a particularly bad or slow.
As for the ALSA-plugin for the libavcodec based resampler, here is their formula:
cutoff = 1.0 - 1.0/filter_size; if (cutoff < 0.80) cutoff = 0.80;
(supposedly in terms of Nyquist frequency)
Libsamplerate uses the following parameters:
Fastest: half-length of the filter is 2464 points, increment 128 (i.e. every 128th point is used for a given convolution), half-period of the sine 154
Medium: half-length of the filter is 22438 points, increment is 491, half-period of the sine is 534
Highest quality: half-length of the filter is 340239, increment is 2381, half-period of the sine is 2465