Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

Semi-Continuous Segmental Probability Model for Speech Signals

Jun Liu, Xiaoyan Zhu, Bin Jia

State Key Laboratory of Intelligent Technology and Systems, Tsinghua University, Beijing, China

A semi-continuous segmental probability model, which can be considered as a special form of continuous mixture segmental probability model with continuous output probability density functions sharing in a mixture Gaussian density codebook, is proposed in this paper. The amount of training data required, as well as the computational complexity of the semi-continuous segmental probability model(SCSPM)[2], can be significantly reduced in comparison with the continuous segmental probability model(CSPM). Parameters of the vector quantization codebook and segmental probability model can be mutually optimized to achieve an optimal model/codebook combination, which leads to a unified modeling approach to vector quantization and segmental probability modeling of speech signals. The experimental results show that the recognition accuracy of the semi-continuous segmental probability model is higher than the semi-continuous hidden Markov model and continuous segmental probability model.

Keywords: hidden Markov model, segmental probability model, semi-continuous segmental probability model, speech recognition


Full Paper

Bibliographic reference.  Liu, Jun / Zhu, Xiaoyan / Jia, Bin (2000): "Semi-continuous segmental probability model for speech signals", In ICSLP-2000, vol.4, 640-643.