[sdiy] Re: Talking about vocal filters....

[jbv]

Hi there,

I might be re-inventing the wheel here (and this is perhaps
of little interest to most of you) but here's what I did yesterday
evening (as there wasn't anything good on TV) :

As I'm still brianstorming on an analog speaking / singing
machine, I decided to make a DSP simulation on MAX-SIM.
I used 3 oscillators, mixed their ouputs and convolved the
output signal with an input signal using 1024 pts FFTs.
The freqs of the 3 oscs are tuned to the 3 first formants of
vocal sounds, and their respective amps are more or less the
requested amp for each formant.

The result are very acceptable vocal sounds (at least for the
basic vowels and with a digital white noise signal as input).
Transitions between vocal sounds are also pretty acceptable
(I used a 100 msec portamento between successive freqs).

OK, this is pure DSP and has nothing to do with analog.
The goal was just to test the values for formants freqs &
amps.
Besides, the result of convolving a white noise with sine osc
through a 1024 FFT gives a similar result as a steep BP
filter with roughly the requested bw for each formant (because
of the 1024 bins).

BTW, I noticed that the amps of each osc doesn't really
matter (it can vary in large proportion without really
destroying the final auditive perception) : freq 1 & 2 can  be
of the same amp, and osc 3 (the higher freq) needs only be
slightly lower than the 2 others...

Well, that's all for now...
Any thought on this ?

BTW, as this doesn't use much of DSP power, I was
wondering if this could be a way to achieve some DSP-based
formant synthesis module...

JB