Research on Tibetan Speech Endpoint Detection Method Based on Extreme Learning Machine

The traditional method of Tibetan speech endpoint detection will reduce the detection accuracy in low SNR environment, so a new method based on limit learning machine is proposed. Firstly, speech signal is preprocessed. By analyzing the human discourse generation model, the speech signal is filtered and processed by frame, the high frequency interference signal in the signal flow is filtered, and the signal flow is decomposed into multiple frames by using the short-term stationary characteristics of the speech signal, so that the characteristics of the speech signals in each frame are kept constant; the sentence analysis is optimized and the word association in the sentence is analyzed by using line graph syntax The system forms an independent semantic block, introduces the pattern template based on pseudo matrix, generates pseudo points through the original feature vector structure and inserts it into the original speech features for protection. Finally, the classification algorithm of limit learning machine is introduced, the optimal configuration of i-h-o is selected, and the end detection method of Tibetan speech based on limit learning machine is completed. The simulation results show that the accuracy of the speech endpoint detection results is significantly higher than that of the traditional method under different SNR and ambient noise.