>> I've been considering a way to have access to frequency information in
>> realtime in csound. Id like to read a window of samples from a file,
>> and then extracting freq info from the written table.

David M. Boothe,
>Take a look at hetro/adsyn for frequency vs. time, if your sound doesn't
>have much transient information. Also look at pvread (uses files
>produced by pvanal).

I know the csound utilities. They let me modulate the signal and timestretch 
it, but I dont have access to the freq value. I wanted to be able to convert 
a window of samples to freq domain (DFT?), and then back.

What I'm looking after is a way to reduce the audio data. I read a sondfile 
in chunks, and try to describe these chuks using a smaller alfabet. The 
samples use a 16 bit=65536 letters alfabet, a bit too much, but I think this 
data is reduced, if instead of the instantaneous values, I use the sample 
variations over time. If for instance I use the difference between 
consecutive samples, I'll be dealing with smaller ranges, and will still be 
able to reconstruct the original signal (I've done it: for instance, 
creating a new soundfile from the sample differences of some other file, you 
can still hear a similar sound to the original, but much quieter). Working 
with difference values, the "dynamic range" is much shorter, and my alphabet 
can be shorter too.

But I want more effective ways. Maybe the spectral ugens can help with this? 
anyone knows of additional documentation/examples than that in the manual?