need a dd-alike for sounf files, running in pipes, say,
(in a stereo file, one sample means a pair of monophonic samples.)
nedd number of bits per sample in header. Format can be packed or
a whole nymber of cwbyteords per monophonic sample.

Huffman differential encoding for transmission over slow line
(eg tty).  Adaptive?  per-block?  Adaptive has almost zero latency.