Newsgroups: comp.dsp,comp.speech
Path: lyra.csx.cam.ac.uk!warwick!uknet!bt!pipex!howland.reston.ans.net!newsserver.jvnc.net!newsserver.egr.uri.edu!kumar
From: kumar@ele.uri.edu (Ramdas Kumaresan)
Subject: Harmonic Partials
Sender: news@egr.uri.edu (USENET News System)
Organization: URI Department of Electrical Engineering
Message-ID: <CrJrD7.CEM@egr.uri.edu>
Date: Fri, 17 Jun 1994 15:07:54 GMT
Lines: 58
Xref: lyra.csx.cam.ac.uk comp.dsp:7716 comp.speech:2690



For the past two years we have been developing techniques and
algorithms for decomposing voiced speech signals into their
harmonic partials with various applications in mind.  Initially
we tried parametric/high-resolution spectrum analysis
based methods for achieving this but they turned out to be
difficult and computationally very expensive. These were described in papers 
given below:

  R.Kumaresan, C.S.ramalingam and A.G.Sadasiv, "On Accurately tracking the
  harmonic components' parameters in voiced speech segments..",  27th 
  Asilomar Conf. on Signals, Systems and
  Computers, Pacific Grove, CA, Nov. 1992, pp. 472-475.

Later we developed tracking procedures for following the frequencies
of individual harmonics. These are discussed in papers given below:

  R. Kumaresan and C.S. Ramalingam, ``On Separating Voiced-Speech into
  its Components,'' 27th Asilomar Conf. on Signals, Systems and
  Computers, Pacific Grove, CA, Nov. 1993, pp. 1041--1046.

  C.S. Ramalingam and R. Kumaresan, ``Voiced-Speech Analysis Based on
  the Residual Interfering Signal Canceler (RISC) Algorithm,'' Proc.
  of IEEE ICASSP-94, Adelaide, Australia, Apr. 1994, vol. I, pp. 473--476.

  R. Kumaresan, C.S. Ramalingam, and A. Rao, ``RISC: An Improved Costas
  Estimator-Predictor Filter Bank for Decomposing Multicomponent Signals,''
  To appear in the Proc. of Seventh SSAP Workshop, Quebec City,
  Jun. 1994.

These methods are improvements of the originally proposed Residual
Signal Analysis (RSA) method of Costas (in 1980) which uses frequency
discriminator loops to track the frequencies of multiple time-varying
sinusoids.  We have applied our method to voiced speech
as well as computer-generated non-harmonically related signals.

One of our main results is that the partials in voiced speech are only
nominally harmonic and not precisely so and that there can exist
brief periods of time in which certain harmonics can become significantly
mistuned.  The envelopes of the harmonics can also exhibit significant
amplitude modulation.

A more complete version of the results will be available in the
Ph.D thesis of C.S. (Ramli) Ramalingam (Department of Electrical
Engineering, Kelley Hall, University of Rhode Island, Kingston, RI 02881)
by the end of summer.  We are also in the process writing a journal
paper that summarizes the results.  Copies of the above mentioned papers
may be obtained by e-mail requests to either me or Ramli (ramli@ele.uri.edu).

Ramdas Kumaresan
Dept. of Electrical Engineering
U. of Rhode Island
Kingston, RI 02881

kumar@ele.uri.edu


