Sistema de conversión texto-voz en lengua gallega basado en la selección combinada de unidades acústicas y prosódicas

In this contribution we describe a corpus-based text-to-speech system for Galician. While traditional concatenative speech-synthesis systems generally employ a quite reduced set of speech units, corpus-based synthesis systems consider many instances of every unit and, by means of dynamic programming techniques, select the sequence of units that minimizes a cost function. With reference to prosody, traditionally, the generation of the prosodic information is carried out in a previous stage to unit selection. This fact implies that, in many cases, the selected speech units must be manipulated in excess in order to fit the desired prosody. In this paper we also propose a method for combined selection of the intonation contour and the sequence of speech units in order to minimize the distortion due to prosodic modifications.