Newsgroups: comp.speech
Path: pavo.csi.cam.ac.uk!warwick!pipex!uunet!elroy.jpl.nasa.gov!ames!koriel!sh.wide!wnoc-kyo!atrwide!atr-la!lucke
From: lucke@atr-la.atr.co.jp (Helmut Lucke)
Subject: Re: Help: a simple speech pattern matching?
In-Reply-To: leeda@cps.msu.edu's message of 13 Jul 1993 23:30:12 GMT
Message-ID: <LUCKE.93Jul14154646@atrq28.atr-la.atr.co.jp>
Sender: news@itl.atr.co.jp (USENET News System)
Nntp-Posting-Host: atrq28
Organization: ATR Interpreting Telephony Research Labs., Japan
References: <21vgi4$v39@msuinfo.cl.msu.edu>
Date: Wed, 14 Jul 1993 06:46:46 GMT
Lines: 35

In article <21vgi4$v39@msuinfo.cl.msu.edu> leeda@cps.msu.edu (Dae Hee Lee) writes:

   >>I have implemented a simple speech pattern matching, but failed. I don't
   >>know what's wrong.  The process was follows:
   >>I recorded hello.wav and goodbye.wav with 11k and 16-bit.  As I
   >>extracted25-channel lpc for each frame.  (I used the frame size 256 and
   >>window size 512.)  I saved in a tablet the parameters frame by frame.
   >>Later I recored another sound file newgoodbye.wav, and processed the
   >>sound in the same way as before and extracted the parameters, and tried
   >>to match it with the hello and goodbye by means of dynamic time warping
   >>(nonlinear method).  But it was misclassified.  So I changed the frame
   >>size and window size, or extracted other parameters such as lpc ceps,
   >>fft ceps, etc. instead, but the result was the same.
   >>Could you point out what is wrong?


Well, it is difficult to know what is going wrong, without knowing
more details but here is my guess:

I think you are making a mistake in the dynamic progamming match
procedure, in that you are not compensating for the different
lengths of your reference words.
Presumably the file hello.wav contains less frames than goodbye.wav.
This presumably leads to fewer comparisons when you match your
input file against `hello' than when you match it against `goodbye'.
Thus the total distance score in the case of `hello' will always
be lower, simply because you are doing fewer local distance computations.

--------------------------------------------------------------------
Helmut Lucke                                <lucke@itl.atr.co.jp>
ATR International Telecommunication Research Laboratories
2-2 Hikaridai, Seika-cho, Soraku-gun, Kyoto 619-02 Japan
Tel: +81-7749-5-1382 (direct)               Fax:   +81-7749-5-1308
     +81-7749-5-1301 (switchboard)
--------------------------------------------------------------------
