Abstract: Endpoint detection is a crucial component in speech recognition system, especially in the noisy environment, the accuracy of endpoint detection has important effect on system performance. This paper proposes an endpoint detection of speech signals based on Subband Based Cepstral（SBC） parameters. Speech signal is measured with wavelet transformation on spectral band, and the energy of SBC is used in speech endpoint detection. Comparison between SBC and MFCC, experimental results show that SBC makes better performance in speech endpoint detection.

Key words: time-frequency analysis, Subband Based Cepstral（SBC）, endpoint detection

摘要： 端点检测是语音识别系统的一个重要组成，尤其是在噪声环境中，其准确性对语音识别系统性能有直接影响。提出了一种基于小波子带倒谱系数（SBC）的语音信号端点检测方法，利用小波变换对频带进行尺度划分，采用小波子带倒谱能量检测语音端点。通过与MFCC的仿真对比以及大量实验分析，小波子带倒谱特征在语音端点检测中具有更好的识别性能。

关键词: 时频分析, 小波子带倒谱系数, 端点检测