ESCA Workshop on Prosody Lund, Sweden September 27-29, 1993

A Numerical Model of Pitch Perception for Short-Duration Vocal Tones: Application to Intonation Analysis

Christophe d'Alessandro

LIMSI - CNRS, Orsay, France

An algorithm for automatic intonation analysis is described. It is based on a two-parameter model of weighted time-averaging with threshold for pitch perception. This model can be considered as a non-linear fitter. In a first stage speech is decomposed into short-duration tonal segments using short-term energy. In a second stage these short-duration tones are analyzed using the numerical model. A set of short (static and dynamic) tones are then obtained, together with their (constant or time-varying) pitches. Stylized F0 contours are reconstructed from this set of tones. Stylized contours are resynthesized, and give synthetic sentences which are perceptually identical with natural sentences.

Full Paper

Bibliographic reference.  d'Alessandro, Christophe (1993): "A numerical model of pitch perception for short-duration vocal tones: application to intonation analysis", In Prosody-1993, 234-237.