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

Intensity Prediction for Speech Synthesis in French

Katarina Bartkova, P. Haffner, Danielle Larreur

France Telecom - CNET Lannion A - TSS/RCP, Lannion - France

The goal of the present study is to predict sound intensity for speech synthesis in French. In order to set up a model for intensity prediction, we first studied intensity variation in natural speech. The data base used was hand segmented and phonetically and syntactically labelled The results of this part of the study were introduced into a rule-based model whose parameters were subsequently optimised using stochastic gradient procedure. Next, a neural network based model was developed and trained using part of the labelled data.

Full Paper

Bibliographic reference.  Bartkova, Katarina / Haffner, P. / Larreur, Danielle (1993): "Intensity prediction for speech synthesis in French", In Prosody-1993, 280-283.