I tried to port AF to NetBSD some time back but failed at the sun device
interface level.  The code was rm'd.

AF subsumes a somewhat older version of sox for some of the audio conversion
functions.

In the past, using sox and AF on a Digital Alpha workstation I found
quantization effects at input truly awful. regular buzzing and clicking
came from the aether into datasteams. I had 44.1KHz 16 bit stereo samples
from a campus radio station digitizer, and wanted to cut 8Khz mono alaw and 
mulaw to ship out on the wire.

Stereo to mono downgrade in SOX was also very weak, as was channel selection
for reasons I didn't understand. I think this has improved. SOX and AF both
provide access to fourier and parametric functions for sound filtering. 
AF has a tcl/tk GUI which is smashing if you want to see spectral views of the 
sounds. Its a very X-like method and if you are familiar with xset then the
aset interface to control physical device settings will not be too foreign.

VAT, Van Jacobsen and Stephen McCannes awsome realtime audio conferencing
tool has now been released as source. It has implementations of PCM/DVI/GSM
and LPC coding some of which is really not aimed at music but have very 
powerful speech compression and conversion functions. INRIA's ivs and
Henning Schulzring (sp?) nevot also have implementations of this kind of code.

NCD's audio tool appears almost useless for real usage on a WAN or more than
trivial usage on the LAN. It lacks any kind of re-synchronization method so
you cannot cope with bufferloss or ordering problems. Again, not relevant
if you just want to mung the do-re-mi's. X consortium was meant to be
trying to get AF and NCDAudio into a common framework with the X sync
extensions. This would be way cool and would permit stuff like SMPTE
timestamped audio and video streams to deliver some lipsynch under loss.

Have fun...

-George

PS does anybody really care about SOX having a tremelo and fuzz function? 
I think they would have done better to cut an API for black-box effects and 
let the masses write the Jimi Hendrix stuff. 

Audio needs a version of NetPBM. not monolithic solutions.