voice synthesizer

[Fraser, Colin J]

> -----Original Message-----
> From: KA4HJH [mailto:ka4hjh at gte.net]
> Sent: 29 April 1999 04:32
> To: synth-diy at mailhost.bpa.nl
> Subject: Re: voice synthesizer
> 
> As I recall (poorly), TI's Linear Predictive Coding wasn't 
> PCM (sampling)
> per se but a kind of analysis/resynthesis. Either it couldn't do an
> arbitrary combination of phonemes, or it could do 
> combinations that it had
> in firmware a lot better.

Linear Predictive Coding is a method for compressing stored digital audio.
I suppose you could look at it as a simple type of analysis/resynthesis.

The recording system 'predicts' each sample in the incoming data stream
(based on previous data), and compares it's prediction with the actual data.
It then stores the *difference* between what it predicted, and the actual
data.

The playback system, based on the same prediction algorithm, also tries to
'predict' what the signal will do next.
It then modifies the actual sample it outputs based on the 'error' that was
calculated and stored by the recording system.

Assuming your prediction alogrithm is good for the type of signals you are
encoding (for example, narrow bandwidth speech sounds), this allows you to
store samples very efficiently.


Colin f