关键词: 端点检测, 短时能量, 短时自相关最大值, 短时过零率

Abstract:

The traditional endpoint detection algorithm based on short-time energy can detect speech endpoint accurately in high SNR environment, but the effect is not satisfactory in low SNR environment. For this purpose, an improved algorithm is proposed based on the maximum value of short-time auto correlation and the zero crossing rate. The short-time autocorrelation maximum is used for valid speech segments area and noise, while the short-time zero crossing rate for the effective detection of the unvoiced signal, thus effectively detecting the endpoint of the speech signal. Experiments show that the improved algorithm reduces the detection error compared to the short time energy algorithm in the low SNR environment, and it can effectively detect the speech endpoint.

Key words: endpoint detection, short-time energy, short time autocorrelation maximum, short-time zero crossing rate