INTERSPEECH 2004 - ICSLP
8th International Conference on Spoken Language Processing

Jeju Island, Korea
October 4-8, 2004

Polynomial Regression Model for Duration Prediction in Mandarin

Yu Hu, Renhua Wang, Lu Sun

University of Science and Technology of China, China

Duration modeling is to establish a mapping relationship between the prosodic environment and the segmental duration engendered in natural speech. In this paper, we first study the effect of prosodic features on segmental duration of neutral utterance in Mandarin by introducing a statistical concept---eta squared, then choose more forceful prosodic features and design interaction quantifying algorithm to study the interaction phenomenon among them, and finally determine the duration model using a polynomial and obtain the coefficients through nonlinear regression. Our research work indicates that 5 to 6 prosodic features might by and large assist a close and accurate mapping between prosodic environment and the perceived duration. Compared to Wagon tree method, this one has undeniable merits.

Full Paper

Bibliographic reference.  Hu, Yu / Wang, Renhua / Sun, Lu (2004): "Polynomial regression model for duration prediction in Mandarin", In INTERSPEECH-2004, 769-772.