Fourth European Conference on Speech Communication and Technology

Madrid, Spain
September 18-21, 1995

Automatic Data-Driven Prosodic Modeling for Text-to-Speech

E. Lopez-Gonzalo, Lúis Hernandez-Gomez

Dep. Senales, Sistemas y Radiocomunicaciones. E.T.S.I. de Telecomunicacion. UPM, Cdad. Universitaria s/n., Madrid, Spain

Prosodic modeling is very important for improved naturalness in text to speech synthesis systems. This paper describes an automatic data-driven methodology for prosodic modeling that can be incorporated into a text-to-speech system. This methodology models both fundamental frequency and suprasegmental duration from a monospeaker recorded corpus. The proposed automatic methodology has the advantage that can be adapted to a specific corpus or a particular speaker. The results of the automatic methodology are compared with our previous manual methodology with the same prosodic data. A greater variability in prosodic contourns are obtained as the main result.

Full Paper

Bibliographic reference.  Lopez-Gonzalo, E. / Hernandez-Gomez, Lúis (1995): "Automatic data-driven prosodic modeling for text-to-speech", In EUROSPEECH-1995, 585-588.