It can certainly be done - I think it is more a question of sound quality|
(intelligibility / speaker recognition / artifacts). At 600bps I doubt you will
achieve speaker recognition.
There are a few papers floating around for waveform codecs around the
1200-600bps range, but the lowest rate standard codec I've heard of is half rate
MELP, which comes in at 1200bps. I don't know of any full implementations which
go lower. From what I've seen so far, waveform codecs seem to be on an
approximate par with LPC based codecs.
If its a commercial project, I would suggest you propose 1200bps or less - which
it totally achievable. If you hit 600bps, all the better - but don't waste too
much engineering time assuming 600bps until you actually hit the rate.