Using GMM for voiced/Voiceless Segmentation and tone Decision in Mandarin Continuous Speech Recognition

摘要

In this paper, methods of Gaussian Mixture Model (GMM) are presented for both silence/voiced/voiceless segmentation and tone decision in mandarin continuous speech recognition system. GMM has been used for silence/voiced/voiceless segmentation before, but the feature parameters can be modified to improve both accuracy and speed. As a popular method in pattern recognition, GMM is first proposed for tone decision. The two GMMs used are proved to be capable and potential.