On Wed, 2015-03-11 at 22:07 +0300, Andrey Semashev wrote: > On Wednesday 11 March 2015 20:48:11 you wrote: > > On Wed, 2015-03-11 at 20:27 +0200, Tanu Kaskinen wrote: > > > > > > I'm writing release notes for 7.0, and I'm wondering how to describe the > > > three soxr resampler variants. Alexander says that all variants are > > > perfect quality-wise (no audible distortions). Alexander also says that > > > each variant takes about the same amount of CPU time, but Andrey says > > > that there's 2x difference between mq and vhq. Who's right? > > > > > > To me it sounds like the hq and vhq variants are redundant, since mq is > > > at least as fast (and on some hardware significantly faster) as the > > > other variants, and there's no meaningful difference in quality. > > > > I now wrote something to the notes, feel free to comment if you'd like > > to change something: > > http://www.freedesktop.org/wiki/Software/PulseAudio/Notes/7.0/ > > Technically, "up to 20 ms" is not correct because sometimes the delay may be > higher. I was careful about that when I updated the man pages. Sorry, when I read "usually up to around 20 ms, in rare cases more" in the man page, I dropped the word "more" in my mind, so I thought 20 ms was a rare case. I'll correct the wiki page. > Regarding hq vs vhq, I think I mentioned it somewhere, vhq does computation > with more precision bits than hq, so it mostly targeted for high bit depth > formats. Whether or not the difference is audible and under what conditions is > an other question. I did some performance measurements, they are given on the > page I referenced earlier [1]. I guess this is mainly a comment on this sentence: "We don't know what evidence the libsoxr developers base their recommendation on." I'll try to choose a more careful wording. Thanks for your input! -- Tanu
