中文摘要：

为了进一步提高低信噪比下语音激活检测（VAD）的准确率,针对基于统计模型的似然比检验方法容易造成非语音信号误检的问题,提出了一种基于子带保留似然比的VAD算法.在似然比综合判决准则中引入保留因子,根据人耳对语音感知敏感度在频率上的差异,采用全局非均匀、局部均匀的方式划分子带,对似然比大于阈值的频率分量所对应的子带信号提取语音特征并根据特征强度设置保留因子,只使用保留的频分量似然比进行综合判决.实验结果表明：该算法在不同的低信噪比环境下检测性能均优于LRT（似然比检验）、MO-LRT（多观测似然比检验）等算法,其漏检率更低,同时有效减少了非语音信号似然比虚高而导致的误检,VAD判决准确率平均提高了2%~14%.

英文摘要：

To improve voice activity detection （VAD） accuracy in low signal noise ratio environments , a robust VAD algorithm based on sub‐band reserved likelihood ratios （LRs） was proposed ,aiming at alleviating the false alarm problem in detection of non‐speech signal of the statistical model likelihood ratios test （LRT ）‐based method .Reserved factor was employed in likelihood ratio decision rule and determined by speech feature strength of sub‐bands ,which were divided global non‐uniformly and lo‐cal uniformly on the basis of human auditory sensing charateristic .The feature was extracted from sub‐band signal correspond to the frequency range ,in which the LRs exceeded certain threshold .The reserved LRs of frequency component were used for final decision .Experiment conducted on various noisy scenarios shows its better performance in comparison with LRT ,MO‐LRT （multiple observa‐tion likelihood ratio tests） ,etc .The clipping rate is lower and the false alarm problem caused by the virtual height of LRs in detection of non‐speech is alleviated ,and the VAD accuracy is increased by an average of 2% ~14% .