基于短时能量比的语音端点检测算法的研究

研究了噪声环境下，利用短时高低频能量比进行语音端点检测的问题。在高信噪比的情况下利用传统的双门限判决算法，在低信噪比的情况下采用短时高低频能量比并辅以过零率为特征参数的算法，保证了在高、低信噪比环境下的端点检测的准确性。试验结果表明，与传统的能量阈值法相比，提出的算法具有更好的性能，是一个简单、高效和稳健的语音端点检测算法。

Speech Endpoint Detection Algorithm Analysis Based on Short-term Energy Ratio

This paper analyzes speech endpoint detection using the ratio of the average energy in the low-bands to that in the high-bands in noisy environments. It employs traditional double-threshold algorithm in the high SNR（Signal-Noise Ratio） environment and uses the ratio of the average energy in the low bands to that in the high- bands and the short-time zero-crossing rate in the low SNR environment, thus to ensure the accuracy in any environment. Experiments show that the proposed algorithm outperforms traditional double-threshold algorithm for speech endpoint detection and proves its accuracy, simplicity and robustness.