Return to the archive index

Re: Timeleine'Puter (was Small Storage Devices

On Mon, 04 May 1998 17:39:05 +0000,  "R. Paul McCarty" <> wrote:

>So you're thinking of basically a generic DSP card that can be setup to
>convert A/D, D/A for whatever you want.  This might work, but it means
>text to speech would have to be coded by hand (or at least somewhat if
>you take advantages of builtin conversion on chip). I think I found a
>couple of speech synthesizer chips that had builtin a/d d/a functions so
>it might be possible to find a chip that has the speech synthesis
>hardwired.  I was thinking more along the lines that I send text to an
>off-board chip and it generates speech (as best it can).

You wouldn't have to hand code the speech, just code for this target.  

While I eventually plan to use festival for speech output, I have had a lot of experience with
rsynth.  rsynth takes text input, converts it to phonemes, converts the phomenes to 50 ms packets
(holmes.c), and sends the packets to a parametric synthesizer (klatt.c).  The parameters are the
formants and fundamental frequency.  At one point I had a DOS port running on a 386DX40 at about 90%
of realtime by rewriting holmes.c and klatt.c to use integer rather than floating point arithmetic.
Profiling showed that > 80% of time was spent in klatt.c

All you would have to do is rewrite klatt.c so that it piped the frames to the dsp, and write
klatt.c for the dsp.  If you use festival, you'd have to rewrite the final stage (isn't it set up
for modular synthesis yo you can use MBROLA, etc?).

>How big is the kit when assembled?

The evaluation kit looks to be about 5x8 inches.

From Wear-Hard Mailing list Archive (WH)
Maintained by R. Paul McCarty

Archive created with babymail