"Stephen D. Williams" wrote:
> Whatever Truespeech calls their 150 bytes-per-second coder. I played
> with true parametric speech years 4+ years ago at AOL. You could play
> back voice on the remote end with voice styles like little girl or
> alien.
At some point, you should do nicely with physical modelling of the human
voicebox (pre-parametrized, beam over you voice metrics prior or during
the conversation), and use speech recognition at the other end. It's hard
to see how this coding should use more than a few baud.
Of course, it goes really hard on the DSP crunch. But we've got more than
a decade of straight-lop-plot Moore with photolitho still, or so I heard.
This archive was generated by hypermail 2b29 : Sun Apr 29 2001 - 20:26:05 PDT