Lennart Augustsson wrote:

> This is probably a reasonable compromise for going in to the
> kernel since it is efficient.  It leads to rather poor audio
> quality though, so we might consider more complicated filters
> in the future.

I don't know much about sampling rate conversion, but since you're using
two constant rates 8kHz <-> 44.1kHz, or 8kHz <-> 48KHz, there might be
away to highly optimize more sophisticated algorhythms. Though I doubt
there's a necessity for setting /dev/audio to 8bit/mulaw/mono. IMHO there
are very few people who use anything but 11025/22050/44100/4800kHz and
many software performance inter/extrapolation by itself.

Just my 0.02 euro cents.