关键词: 端点检测, 频谱方差, 谱减法, 语音识别

Abstract: In order to improve correctness of endpoint detection method based on spectrum variance in the case of low Signal Noise Rate（SNR）, this paper proposes a new speech endpoint detection algorithm based on combination spectrum variance with spectral subtraction. It reduces speech signal noise dynamically by using modified spectral subtraction and calculates its variance to set double-threshold in endpoint detection. Simulation results indicate that the proposed algorithm has advantages of high noise-robust and adaptive performance. Furthermore, it has better detection capability than spectrum variance method in the case of low SNR.

Key words: endpoint detection, spectrum variance, spectral subtraction, speech recognition