[Cuis-dev] Text to Speech (was Re: Porting State of Magma)

[Luciano Notarfrancesco]

On Wed, 9 Jun 2021 at 4:36 PM Gerald Klix via Cuis-dev <
cuis-dev at lists.cuis.st> wrote:

>
>
> On 6/8/21 10:11 AM, Luciano Notarfrancesco via Cuis-dev wrote:
> > I haven’t used image segments in many years too, and I don’t think I’ll
> >   need them for anything in the foreseeable future. In the past I used
> them
> > twice, as far as I can remember. The first time it was to store diphones
> > (pairs of phonemes) for a text to speech system that was much better than
> > the one that ended up in the Squeak image, but had the disadvantage of
> > depending on this database of diphones.


> Do have this one still around? Does it support German, or can it be
> changed to support German?


It was around 1998, I lost it. It used the diphones sets from Festival, and
I never tried it with German but I think Festival does have a German voice.
The synthesis method was residual-excited LPC synthesis, pretty easy to
implement in a few lines of code (at the time I made a plugin in slang, but
perhaps it can run well in pure Smalltalk in modern computers). I guess the
hardest part when adding support for a new language is the text to phonemes
conversion, once you have phonemes and an intonation contour (if I remember
correctly) the conversion to diphones and synthesis are language
independent.

I’ll try to find it on the internet, but I’m pretty sure it’s lost…

Best regards,
Luciano

>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cuis.st/mailman/archives/cuis-dev/attachments/20210610/6b558a8f/attachment-0001.htm>