Newsgroups: comp.speech
Path: pavo.csi.cam.ac.uk!warwick!pipex!uunet!munnari.oz.au!cs.mu.OZ.AU!miscord
From: miscord@ee.mu.OZ.AU (Michael Scordilis)
Subject: Re: Algorithm for pitch detection
Message-ID: <9314418.13015@mulga.cs.mu.OZ.AU>
Sender: news@cs.mu.OZ.AU
Organization: Department of Electrical & Electronic Enginnering, University of Melbourne
References: <1993May18.162011.3214@Princeton.EDU> <1993May18.184742@IASTATE.EDU> <C79F11.C39@rahul.net>
Date: Mon, 24 May 1993 08:08:00 GMT
Lines: 32


Pitch Detection Algorithms (PDA) with accuracies of up to 3% are things of
the past.  In fact, pitch estimates within small a fraction (1/8, 1/16, etc) 
of a sampling period is possible.  Now, bearing in mind that a digital 
signal is not defined between sample values, this is significant.
Very accurate pitch detection schemes are those based on cross-correlation
criteria on either the time or the frequency domain. 

For a time-domain algorithm, you can check:

Y. Medan, E. Yair, D. Chazan, "Super Resolution Pitch Determination of
Speech Signals", IEEE Trans. Sing. Proc., Vol. 39, No.1, Jan. 1991, 
pp. 40-48.

For a frequency-domain algorithm, you can check our algorithm: 

T S Lim, M S Scordilis, "New and Improved Pitch Determination for the 
IMBE Vocoder", Proceedings of the Fourth Australian International 
Conference on Speech Science and Technology, Brisbane, Dec 1992, 
pp. 375-380.



Now, on non-harmonically related pitch determination, I am not sure
as to whether you could observe or track such a quantity, since there
will be no dominant temporal periodicity in such a signal.  Basically,
I don't think such thing exists.  I cannot imagine it. 
Perhaps one could provide some examples (description or samples).



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