基于最大熵谱估计和时频特性的语音端点检测  

SPEECH SIGNAL ENDPOINT DETECTION BASED ON MAXIMUM ENTROPY SPECTRUM ESTIMATION AND TIME-FREQUENCY SIGNATURE

作　　者：陈莹莹 简磊 

机构地区：[1]四川大学锦江学院电气与电子信息工程学院,四川彭山620860

出　　处：《计算机应用与软件》2017年第11期91-96,共6页Computer Applications and Software

摘　　要：语音端点检测对于构建实际语音识别系统具有重要的意义。为了提升在低信噪比条件下语音端点检测算法的性能,提出一种基于最大熵谱和时频特性的端点检测算法。对分帧后的语音信号通过最大熵估算出功率谱,并根据带噪语音信号时频域上的特性进行特征捕捉,从而进行端点检测。实验结果表明,此方法在较低的信噪比下(-9~0 dB)能够比较准确地捕捉语音信号的特征,明显地提高了端点检测的准确性。Speech endpoint detection is crucial to the construction of a practical automatic speech recognition system. A new algorithm based on the maximum entropy spectrum estimation and time-frequency signature is proposed to improve the performance of speech endpoint detection in low SNR （Signal Noise Ratio） environment. The framed speech signal power spectrum was estimated through the maximum entropy, and then the characteristics of noisy speech were extracted in time-frequency field in order to detect the endpoint. Experimental results show that, this method can accurately capture the characteristics of speech signals under lower SNR （ -9 - 0 dB）, and significantly improves the accuracy of endpoint detection.

关 键 词：端点检测 最大熵谱估计 时频特性 信噪比 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]