EUROSPEECH 2003 - INTERSPEECH 2003
8th European Conference on Speech Communication and Technology

Geneva, Switzerland
September 1-4, 2003

        

Segmental Durations Predicted with a Neural Network

Joao Paulo Teixeira (1), Diamantino Freitas (2)

(1) Polytechnic Institute of Braganca, Portugal
(2) University of Porto, Portugal

This paper presents a segmental durations' model applied to the European Portuguese language for TTS purposes. The model is based on a feed-forward neural network, trained with a back-propagation algorithm, and has as input a set of phonological and contextual features, automatically extracted from the text. The relative importance of each feature, concerning the correlation with segmental durations and improvements in the performance of the model, is presented. Finally the model is evaluated objectively and subjectively by a perceptual test.

Full Paper

Bibliographic reference.  Teixeira, Joao Paulo / Freitas, Diamantino (2003): "Segmental durations predicted with a neural network", In EUROSPEECH-2003, 169-172.