Fourth European Conference on Speech Communication and Technology

Madrid, Spain
September 18-21, 1995

High-Quality Speech Modification based on a Harmonic + Noise Model

Yannis Stylianou, Jean Laroche, Eric Moulines

Ecole Nationale Superieure des Telecommunications, Departement Signal / CNRS-URA 820, Paris, France

In this contribution a novel method enabling high-quality speech modifications is presented; it is based on a harmonic + noise, (HNM), representation of the speech signal. The harmonic part is modelled by harmonically related sinewaves with slowly varying amplitudes, and frequencies. The noise part is modelled by an AR model and it is modulated by a time-domain amplitude envelope. Both the analysis and the synthesis are pitch-synchronous. Thanks to this twofold representation (Harmonic + Noise part), and to the pitch-synchronous scheme, a flexible technique for time-scale and pitch-scale modifications, inspired by PSOLA methods, can be applied, yielding high-quality modification.

Full Paper

Bibliographic reference.  Stylianou, Yannis / Laroche, Jean / Moulines, Eric (1995): "High-quality speech modification based on a harmonic + noise model", In EUROSPEECH-1995, 451-454.