Research On Endpoint Detection Algorithms Of Speech Signal

Speech endpoint detection, which accurately detects the beginning and ending of the speech signal from background noise, is crucial for the speech signal processing. The accurate endpoint detection can improve the speech recognition accuracy with less recognition time, and decrease the noisy bit rate. Moreover, the efficiency of the speech coding can be increased by endpoint detection. Generally speaking, the performance of the detection severely degrades in the actual noisy condition. Therefore, the speech endpoint detection is particularly important in the speech recognition, especially for low SNR. Firstly, with the analysis of some speech features, the existing typical algorithms of endpoint detection are summarized. And some simulation results and improvements are also shown in the paper. Then, two new algorithms of the speech endpoint detection are proposed. The first one: based on the hearing system of the human, a new adaptive time frequency parameter considers both auditory and vocal tract character is proposed by studying the Mel filter bank. Simulation results have shown the performance of the algorithm is better than other existing algorithms. The second one: combined with the Mel cepstrum distance and the multi-band energy entropy, animproved isolated word endpoint detection algorithm is proposed, which can exempt the adjustment of the decision threshold by means of estimating background noise. Simulation results show the algorithm has better robust capability , and high efficiency can be obtained in common noisy environments. At the end of the article, the two proposed algorithms are summarized. Some new study fields within the past two years are introduced and developing perspective of endpoint detection is referred to.