|
|
|
| Study on Fundamental Frequency Model Based on MMSE Principle |
| Liu Hao-Jie; Du Li-min |
| SITR, Institute of Acoustics, Chinese Academy of Sciences, Beijing 100080, China |
|
|
|
|
Abstract Based on the interaction of tone and intonation, a new method is proposed to get the top line and the bottom line for continuous speech of mandarin, using the Minimum Mean Square Error (MMSE) principle. The research results demonstrate that the top line and the bottom line have different function to the intonation of continuous speech on the part of quantum. The paper also proposes a primary mathematical model to figure the interaction between the tone and intonation. On the basis of the extraction of two lines and the primary model, the two-lines model and the median fundamental frequency model are established separately, which provides a flexible and effective method to modify the intonation and improves the naturalness of speech synthesis system of mandarin greatly.
|
|
Received: 24 July 2004
|
|
|
|
|
|
|
|
|
2005, Vol. 27 
