[MUD-Dev] Free FFT

[Matthew Estes]

From: Christopher Kohnert <chris at achaea.com>

> Seems a lot of people are pretty knowledgeable about this voice
> trickery. I'm interested in doing some research with voice, but
> need to either implement my own FFT or find a freely available
> package tailored to voice analysis. (The 'front end' of a speech
> recognition perhaps.) My question is: do any of you know of such a
> package? I'm not really interested in doing speech recognition,
> but I am interested in FFT'ing a voice to reduce the amount of
> data analysis required.

I don't know of any packages off-hand, but as a senior Electrical
Engineering student who has had several courses in Signal
Analysis(and a Digital Signal Processing course) I can tell you a
couple of things. FFT won't reduce the size of the "signal" you
have. It will just transform it from the "Time" domain, to the
"Frequency" domain(it gives you those nice little spectrums displays
you see in Winamp and the ilk).

Roughly(I know there are LOTS of details), the way you then compress
it is to apply "psychoacoustical" models of human hearing to
eliminate frequencies the human ear can't hear or won't notice. This
will produce a LOT of zeros in your signal. Then you use normal
compression techniques to compress your decimated signal.

Really, if you don't want to do a lot of math or spend several
months working on something like this(years to get it REALLY good),
you're better off finding someone who has already wrote a
compression package for realtime voice transmission.
_______________________________________________
MUD-Dev mailing list
MUD-Dev at kanga.nu
https://www.kanga.nu/lists/listinfo/mud-dev