[sdiy] Texas Instruments LPC Speech Chips

[Gordonjcp]

On Tue, Sep 01, 2026 at 11:18:38AM +0100, Tom Wiltshire wrote:
> Certainly there's lots of stuff out there on Linear Predictive Coding, the compression method TI used to get (vaguely!) intelligible speech into the tiny amount of ROM they had available. Beyond that, the LPC model could be done with analogue tech - a "source" oscillator representing the vocal cords, a series of formant filters to create the resonances of the throat and mouth, and then perhaps something to create noise and pops for plosives and such. The trouble is getting hold of the required data to *control* such a thing and make it produce something that sounds like talking. That problem would remain even if you did the entire production chain digitally. Or perhaps I'm wrong and there's loads of LPC datasets freely available now?

It's easy enough to encode and decode LPC on PCs and indeed there is a codec for sox that does it. When I record a sample of speech, LPC it, and then un-LPC it, it sounds like a Speak'n'Spell with my accent - very little trace of my voice's timbre remains but the cadence and articulation is exactly me.

It's really quite strange. I recommend you try it.

-- 
Gordonjcp