7th International Conference on Spoken Language Processing

September 16-20, 2002
Denver, Colorado, USA

Application of Microprosody Models in Text to Speech Synthesis

Phuay Hui Low, Saeed Vaseghi

Brunel University, U.K.

This paper presents a Markovian model of the sequential dependency of the acoustic correlates of speech, namely the pitch trajectory, formants trajectories, the power trajectory, the expected duration of successive phonetic speech segments and the speaking rate for text to speech synthesis. Using a first-order Markov model, microprosody is modeled within biphone segments. Since there is a substantial degree of overlap between biphones and syllables, the method can be considered as an effective model for microprosody within syllables. The model is used to improve the quality of the output of a text to speech synthesis system.

Full Paper

Bibliographic reference.  Low, Phuay Hui / Vaseghi, Saeed (2002): "Application of microprosody models in text to speech synthesis", In ICSLP-2002, 2413-2416.