Newsgroups: comp.ai.neural-nets,comp.speech
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!fas-news.harvard.edu!newspump.wustl.edu!news.ecn.bgu.edu!vixen.cso.uiuc.edu!howland.reston.ans.net!Germany.EU.net!ieunet!curia!csws3.ucc.ie!turlough
From: turlough@odyssey.ucc.ie (Turlough FitzPatrick)
Subject: tdnn training
Message-ID: <DJHCGr.2K3@curia.ucc.ie>
Sender: usenet@curia.ucc.ie
Reply-To: turlough@odyssey.ucc.ie
Organization: Computer Science Dept., University College, Cork, Ireland
Date: Tue, 12 Dec 1995 15:31:39 GMT
Lines: 18
Xref: glinda.oz.cs.cmu.edu comp.ai.neural-nets:28613 comp.speech:7761

I am currently trying to train a time-delay neural network.
I have been through the code a couple of times and I'm 
pretty certain that its all functioning right. I am presenting speech 
input to the network that has been windowed, fft'd and mel
cepstrums have been calculated. I tried the network with 
both normalised and un-normalised inputs but there was no 
difference.
The problem is that the network is not returning the correct
output. I constantly get one output node firing strongest regardless
of the pattern presented. Even though there seams to be a reduction
at the end this neurode is recognised on each of the test patterns.

Has anybody experienced this type of problem before or does anybody
have any idea about what could be wrong.
I would appreciate any responce. Thanks in advance,
turlough


