best approach towards speech analysis?

[linked_void]

i've been asked to help on the audio front with a friend's project. the requirement is to have a program which accepts an audio file - in this case a speech, and have it return a directory of files, one for each word of the input file.  i'm wondering how others may have approached this task and what, if any, classes are out there to use. any info greatly appreciated, thanks all!

[dewdrop_world]

I don't remember any discussions on the mailing list about speech recognition. It's a very interesting topic for me -- because of chronic pain problems, I've used dictation software for years (in fact, I'm writing this using iListen from MacSpeech) -- but I haven't had time to explore it myself.

This thread on the electro-music.com forum has some useful pointers -- that guy is using chuck rather than SuperCollider, but the concepts are valid no matter what language you use.

http://electro-music.com/forum/viewtopic.php?t=21647

A lot of audio analysis techniques convert frequency data to the Mel scale, because raw FFT data over emphasize high frequency content. Recently the MFCC UGen was created to produce "mel-frequency cepstral coefficients" from a basic FFT analysis. You might find this very useful.

Be patient with yourself -- this is hard, hard stuff to wrap your head around. I hope these are useful pointers in any case.

hjh