[pulseaudio-discuss] always resampling to float32le?

[Lennart Poettering]

On Wed, 05.08.09 20:11, sladivadi (sladivadi at verat.net) wrote:

> Hi,
> 
> I wonder what sample format I should use in order to minimize resampling
> across various module-combine and module-remap sinks.
> I specified rate=s32le for all virtual sinks I use since my hardware seems
> to only open like this when specifiyong a 32bit or 24bit format. 24bit is
> what my hardware actually supports (Realtek RC888).

Very often 24bit means the 24 MSB of 32bit integer words, which PA
then exposes as s32 NE. How many of those bits are actually used are
exported as properties on the sink device however. (but even that is
not really that reliable, since DAC's generally cannot do more than
20-22 bit. Thre's a physical limit somewhere in that area)

> However when looking at output of pulseaudio -v I see:
> I: resampler.c: Using float32le as working format.
> 
> Is PS resampling two times from for example 16bit to s32le to float32le?

The speex resampler can only do 16bit integer NE and 32bit float
NE. That means we need to convert to either of these two formats
before doing the resampling step.

Lennart

-- 
Lennart Poettering                        Red Hat, Inc.
lennart [at] poettering [dot] net
http://0pointer.net/lennart/           GnuPG 0x1A015CC4