Newsgroups: comp.speech
Path: pavo.csi.cam.ac.uk!doc.ic.ac.uk!agate!spool.mu.edu!wupost!waikato!aukuni.ac.nz!cs18.cs.aukuni.ac.nz!pgut1
From: gouri@eeserv.ee.umanitoba.ca (Srigouri Kamarsu)
Subject: Question regarding DCT
Sender: pgut1@cs.aukuni.ac.nz (Peter Gutmann)
Organization: U of Manitoba, Winnipeg, Manitoba, Canada
Date: Fri, 26 Nov 1993 11:43:38 GMT
Message-ID: <1993Nov26.114338.24169@cs.aukuni.ac.nz>
Lines: 26


HI,
 
I am looking into using Neural Nets for SPeech compression.
I am thinking of doing a DCT on the residual signal after LPC
analysis and VQ the DCT coefficients using Kohenen's learning or
Frequency sensitive competive learning (Krishnamurhty et.al).
 
Right now I have used NN's to VQ the
LPC co-efficients but for the residual, I want to use DCT's. There
was an article in Electronics letters by Velva et.al. in June'93
(if I remember right) who have done
somthing similar but they have used the actual speech signal
itself.
 
I wanted to know if anyone has done something like this and
what the pro's and con's are for using this technique.
Any pointers and references also would be appreciated.
Please e-mail me the responses.
Thanks.
 
--
SriGouri Kamarsu
Dept. of Elec. and Computer Engg.
15 Gillson Street, University of Manitoba,
Winnipeg, Canada R3T 5V6
