Newsgroups: comp.speech
Path: pavo.csi.cam.ac.uk!warwick!uknet!pipex!uunet!noc.near.net!howland.reston.ans.net!newsserver.jvnc.net!newsserver.egr.uri.edu!ramli
From: ramli@ele.uri.edu (Ramli)
Subject: Re: Algorithm for pitch detection
Message-ID: <C7A1I1.Jtx@egr.uri.edu>
Sender: news@egr.uri.edu (USENET News System)
Organization: URI Department of Electrical Engineering
References: <1993May17.200530.16738@en.ecn.purdue.edu> <1993May18.162011.3214@Princeton.EDU>
Date: Wed, 19 May 1993 13:58:49 GMT
Lines: 32

L. Seltzer (lseltzer@phoenix.Princeton.EDU) writes:

>There is a short article by Ken Steiglitz in an IEEE Transactions
>on Acoustics, Speech and Signal Processing - I think it's from the
>late 1970's, where he describes an optimal f0 detection algorithm
>using trigonometric curve fitting. 

The complete reference for the above paper is:

"Pitch extraction by trigonometric curve fitting" by
K. Steiglitz, G. Winham, and J. Petzinger
IEEE Trans. Acoust., Speech, and Sig. Process., 
June, 1975, pp. 321--323.

Methinks this paper is not as well-known as it should be.
I am no expert on pitch estimation and its associated literature
but this paper seems to be among the first ones that I know of that fits
a harmonic model in the least-squares sense to estimate F0.
I, too, am applying a similar technique but became aware of this work
only very recently.  The solution of this estimation by Steiglitz et al.
was not as intuitively appealing to me as the form in which this problem
cast in the spectral analysis literature, i.e., using
projection matrices, and the elimination of the dependence on the amplitudes
by using the classical trick of Golub & Peryera.  The pitch is finally estimated
by using a gradient descent approach, as opposed to the grid search
method used in the above quoted paper.  If the assumption of the harmonic
model is valid over the block size chosen, one cannot do better than
this estimation method, I think.

Ramli.


