TORRES Humberto Maximiliano
congresos y reuniones científicas
Parameter estimation and prediction from text for a superpositional intonation model
HUMBERTO MAXIMILIANO TORRES; JORGE ALBERTO GURLEKIAN
Conferencia; 20 Konferenz Elektronische Sprachsignalverarbeitung ( ESSV 2009); 2009
TU Dresden, Intitut für Akustik und Sprachkommunikation
This work presents an approach for parameter estimation and prediction of the Fujisaki model for Argentine Spanish. Language hypotheses were proposed for estimation and tested by means of genetic algorithms. These hypotheses were validated by comparison of the estimation performance relative to the standard method. Prediction was then calculated based on input text information and performed by CARTs reached a performance comparable to approximations presented for Japanese and English. Objective measures showed that the predicted F0 contours are not close copies of the originals, nevertheless perceptual judgments revealed an impression of good quality for the intonation which makes the method a valuable tool for the development of TTS systems.