Research on Endpoint Detection for Mongolian Speech Based on Support Vector Machine

Endpoint detection is the key technology in system of speech identification. As an object of experiment and research, Mongolian speech attracts more and more researchers. It is significant for the development of Mongolian speech identification technology to apply endpoint detection to Mongolian speech. Support Vector Machine (SVM) is a kind of new technology in the field of Data Mining, this paper applies C-SVM to endpoint detection for Mongolian speech, and overcomes some trivial problems and inaccuracy provoked by setting threshold. Through the experiment on Mongolian speech signal, short-time energy, short-time average zero-crossing ratio and Mel-Frequency Cepstrum Coefficient (MFCC) are extracted as features, and their distinguishing abilities between speech audio segment and non-audio segment is researched with the effect is excellent.