[DECtalk] DECtalk source compiled

[Don]

On 6/22/2020 6:42 PM, Jake Gross wrote:
> It was more like a leak, since Edward Bruckert sent me the code, and I
> preserved it on my sight.

Ah, OK.  So, it is still, technically, encumbered by copyright.

> I also found different implementations of the vocal tract model in the
> code, and I noticed if you change to a different implementation, it
> changes the frequency response of the voice.

Yes.  There are different ways to implement resonators and antiresonators.
In a perfect world, they produce the same results.  But, when constrained
by the chosen capabilities of the software, they exhibit different
characteristics.  In some forms, the implementation may be susceptible
to overflow errors (often not even aware of these possibilities).  In
others, rounding errors distort the signals.

For example, vowel sounds are easily FAITHFULLY produced by a series
configuration of formants.  But, this usually requires more dynamic
range in the math than an equivalent PARALLEL implementation.

Additionally, changing the coefficients (as the model "speaks") introduces
audible artifacts into the waveform.  I noticed in the earliest published
versions of Klatt's synthesizer that he hadn't taken any measures to eliminate
these.  When you consider it was a doctoral thesis, it's easy to imagine that
he was just trying to get something that SOUNDED like it was working (in the
time constraints of an academic calendar)

> I am also going to look into disabling hlsyn.