Newsgroups: comp.speech
Path: pavo.csi.cam.ac.uk!warwick!pipex!uunet!cs.utexas.edu!swrinde!network.ucsd.edu!munnari.oz.au!cs.mu.OZ.AU!miscord
From: miscord@ee.mu.OZ.AU (Michael Scordilis)
Subject: Re: computing speech spectrograms?
Message-ID: <9314616.12886@mulga.cs.mu.OZ.AU>
Sender: news@cs.mu.OZ.AU
Organization: Department of Electrical & Electronic Enginnering, University of Melbourne
References: <C7K30K.LKu.1@cs.cmu.edu>
Date: Wed, 26 May 1993 06:31:21 GMT
Lines: 22

In article <C7K30K.LKu.1@cs.cmu.edu> mkant+@cs.cmu.edu (Mark Kantrowitz) writes:
>
>My input is ulaw-encoded speech files (captured using a mike attached
>to a Sparc) which I've converted to linear. I pass 64 (or 128)
>datapoints at a time to a FFT routine (shifting each time by 16
>points), which returns 64 complex numbers. I interpret the magnitude
>of the ith complex number as the intensity of the sr*i/64 Hz
>frequency, where sr is the sampling rate. I plot only those
>frequencies whose intensity is greater than some appropriate threshold
>value. The resulting plots do not look like spectrograms.
>
>Any suggestions?

I don't know what you mean by "weird results", but you only need to 
plot the first N/2 magnitude points (symmetry in DFT of real sequences).
Check the FFT of a nice vowel portion (e.g., /a/ or  /i/).  Try Hamming
or Gaussian windows to segment your data prior to FFT. 

iMichael Scordilis


Michael Scordilis	   	|internet    miscord@mullian.ee.mu.OZ.AU
