Interspeech'2005 - Eurospeech
Data scarcity in corpus-based intonation modelling for TTS applications is addressed. We propose to apply a searching process to a list of dictionaries of classes of intonation patterns previously trained from corpus to avoid problems associated with the scarce number of samples in the classes. Results indicate that better results are obtained in comparison with previous alternatives where the probability of predicting a less representative intonation pattern has been shown to be higher.
Bibliographic reference. Escudero, David / Cardeñoso-Payo, Valentín (2005): "Optimized selection of intonation dictionaries in corpus based intonation modelling", In INTERSPEECH-2005, 3261-3264.