[pulseaudio-discuss] Best Quality resampling method?

[Richard Shaw]

On Sat, Apr 11, 2009 at 7:12 AM, Mark Greenwood <fatgerman at ntlworld.com> wrote:
> On Saturday 11 April 2009 01:01:04 Lennart Poettering wrote:
>> On Tue, 07.04.09 21:17, Mark Greenwood (fatgerman at ntlworld.com) wrote:
>>
>> > I know this may be subjective, but what is the best quality
>> > resampling method? I'm referring to the resample-method= option in
>> > /etc/pulse/daemon.conf
>> >
>> > Mine is currently set to src-linear but I get the impression from
>> > the docs that speex-float-9 should be better. As it's quite tricky
>> > to do back-to-back listening tests I was wondering if there is some
>> > definite objective opinion on the subject?
>>
>> Depends on your metric do define to be 'best quality'.
>>
>> Closest to mathematically perfect is src-sinc-best-quality.
>>
>> src-linear is one of the crappiest. It uses linear interpolation which
>> is evil shit. If you consider that good quality you have a very, uh, ...
>> distuingished taste. ;-)
>
> Lol, thanks Lennart. src-linear was not chosen by me but by a package maintainer for a distribution that probably ought to remain nameless.
>
> I've switched distros now anyway, for various reasons, but I'll try src-sink-best-quality and see if I can hear the difference :)
>
> Mark
>
>>
>> Lennart

Well I don't know about hearing the difference, as with anything like
that there is probably a diminishing return. I tried
src-sink-best-quality which used 60% CPU in top on one core (Athlon 64
X2 2.7GHz). I didn't get to try it very long because it appeared to
cause mythTV (mythmusic) to crash. What would be one step down from
src-sink-best-quality?

Richard