为提高语音端点检测系统在低信噪(0 dB 以下) 下

检测的准确率, 提出了一种基于谱熵的端点检测算法。将每

帧信号分为16 个子带, 选取频谱分布在250～ 3. 5 kHz 并且

能量不超过该帧总能量90 的子带, 计算经过语音增强后的

子带能量以及各子带信噪比, 根据各子带信噪比的不同调整

其在整个谱熵计算过程中的权重, 然后平滑谱熵, 以最终的

谱熵作为端点检测的依据-To improve endpoint detection system in the low signal to noise (0 dB or less) under the detection accuracy, a spectrum entropy-based endpoint detection algorithm. Each frame signal is divided into 16 sub-bands, select the frequency distribution in the 250 ~ 3. 5 kHz and the energy of not more than 90　of the total energy of the frame of the sub-band, calculated through the following sub-band speech enhancement as well as the sub-band signal to noise ratio of energy , according to the different sub-band signal to noise ratio to adjust its calculation of the spectral entropy of the process of weight, and smooth spectral entropy, spectral entropy to the final endpoint detection as the basis for