[DECtalk] Dectalk Digest, Vol 11, Issue 14

[Tom Kopec]

The modeling of voices looks like it should be simple, but tiny details 
make a huge difference.

Ed was there in the early days, I arrived much later.. but if I recall 
there was some work in live vocal-tract measurements and other 
physically-based methods that were used in creating the original voices.

As Ed mentioned, one of the issues that crops up when you modify a voice 
is the overload problem (running out of the dynamic range of the integer 
DSP code due to unfortunate parameter choices) .. so one might say 
"well, we have infinite compute power these days, just make it floating 
point!" .. turns out that brings up a whole new pile of problems..

Unfortunately, as far as I can tell, the DECtalk code remains 
proprietary.. there's a lot that could be done by folks who really know 
what's under the hood.. and at least some of us would really like to do 
it..  but we can't touch it.

...tom


On 6/27/2015 12:00 PM, dectalk-request at bluegrasspals.com wrote:
> Date: Fri, 26 Jun 2015 15:52:25 -0400
> From: Edward Bruckert <edbruckert at gmail.com>
> To: DECtalk <dectalk at bluegrasspals.com>
> Subject: Re: [DECtalk] a question about something in the
> 	dectalkdocumentation
> Message-ID: <FC288C03-88EB-4DA4-884B-B1E71E1E5B39 at gmail.com>
> Content-Type: text/plain; charset=us-ascii
>
> There are only two data bases in fectalk. All others are vreted using the speaker def parameters. the doctor dennis voice and
> the betty are the two voices model directly modeled after real voices. The earliest dectalks only
> had dthe dennis database. The female was done by badly scaling the malle ones .
> so play with the parameters and create your own, Warning the biggest problem is overloading a filter bank so playing with gains is often necessary.
> On Jun 26, 2015, at 10:28 AM, Blake Roberts via Dectalk <dectalk at bluegrasspals.com> wrote:
>