8th European Conference on Speech Communication and Technology

Geneva, Switzerland
September 1-4, 2003


A New Method for Pitch Prediction from Spectral Envelope and its Application in Voice Conversion

Taoufik En-Najjary (1), Olivier Rosec (1), Thierry Chonavel (2)

(1) France Telecom R&D, France
(2) ENST Bretagne, France

This paper deals with the estimation of pitch from only spectral envelope information. The proposed method uses a Gaussian Mixture Model (GMM) to characterize the joint distribution of the spectral envelope parameters and pitch-normalized values. During the learning stage, the model parameters are estimated by means of the EM algorithm. Then, a regression is made which enables the determination of a pitch prediction function from spectral envelope coefficients. Some results are presented which show the accuracy of the proposed method in terms of pitch prediction. Finally, the application of this method in a voice conversion system is described.

Full Paper

Bibliographic reference.  En-Najjary, Taoufik / Rosec, Olivier / Chonavel, Thierry (2003): "A new method for pitch prediction from spectral envelope and its application in voice conversion", In EUROSPEECH-2003, 1753-1756.