Dectalk 5

[Doug Sutherland]

Cheryl said:

 > You know, it seems to me that the difference is more in the tone
 > than in the actual voice per se.

The IBM via voice engine, which is really an eloquence engine, allows
these settings to be tweaked via their API:

Gender (male/female)
HeadSize (0-100)
PitchBaseline (0-422)
PitchFluxuation (0-100)
Roughness (0-100)
Breathiness (0-100)

Along with other usual stuff like speed etc.

So a "voice" is a specific setting of all those parameters.
Viavoice has 5 pre-programmed voices but you can make your own.

It seems like different voice engines/synths sound different though.
For example, I think festival sounds horrible compared to viavoice,
and I think viavoice sounds better than dectalk.

Is it the "voice" or "engine" that makes it sound better?
I'm not really sure. It doesn't seem like festival is capable of
sounding like viavoice.

I did some work on drivers for a doubletalk eval board. The voice
there is produced by one tiny chip. It sounds pretty much just like
the dectalk software voices. They are decent, but I'd like to find
one that sounds really good. Maybe speech tech still has a ways to
go in its evolution ...

   -- Doug