[DECtalk] DECTalk 4.2 Alpha (fwd)

[Don]

On 8/17/2022 3:33 PM, Brandon Tyson wrote:
> I wonder if this is why the old Braille n Speak always sounded monotone?
>
> Was it because they didn’t take time to make the pitch be more expressive in
> sentences?
>
> The only shift in pitch I tended to notice was at periods and question marks but
> that was it.

The "poor man's" inflection control is "down at full stops, up at
question marks".

[The poorest man uses none at all!  <grin>]

Beyond that, you (the synthesizer) really need to understand sentence
structure.

A more natural approach deals with breath groups and gradually
adjusts pitch/emphasis over them within the sentence (assuming
the text DOES represent a valid sentence; all bets are off
if it's just a collection of words that don't adhere to
grammatical rules)

The original MITalk had a set of low frequency modulators that
tried to make the speech have some sense of variation -- as if
a human couldn't maintain a monotone.

Making waveforms is relatively trivial and takes very little code.
Like one sheet of paper to implement the entire vocal tract model!

Most of the code in a TTS is concerned with trying to understand
the intent of the text presented so that it can emit meaningful
speech.  And, English is notorious for oddball pronunciations...