一种改进的基于高阶统计分析的语音激活检测算法  

Modified Voice Activation Detection Algorithm based on Higher-Order Statistical Analysis

作　　者：孔德廷 KONG De-ting(Southwest China Institute of Electronic Technology,Chengdu Sichuan 610036,China)

机构地区：[1]中国西南电子技术研究所,四川成都610036

出　　处：《通信技术》2020年第7期1699-1703,共5页Communications Technology

摘　　要：提出了一种基于高阶统计的改进型语音激活检测算法。相对于传统的语音激活检测算法,在语音信号存在不确定的条件下,利用LPC残差域的高阶统计属性量,作为语音/噪音信号的判决标准。综合引入的归一化偏度/峰度、偏峰比等高阶统计量和传统的噪音信号概率等一阶分析量,形成语音/噪音软判决机制,能够有效的识别带噪语音信号中的语音激活区域。计算机仿真结果表明,输入背景噪音为高斯噪音和类高斯噪音时,所提算法相对于传统算法的误判概率要低,且在低信噪比条件下,具有更好的识别性能。An modified voice activation detection algorithm is proposed for distinguishing speech from noise.Compared with the traditional voice activation detection algorithm,under the condition of uncertain speech signal,the high-order statistical attribute of LPC residual domain is used as the decision standard of speech/noise signal.Combining the high-order statistics such as normalized skewness/kurtosis and ratio of skewness-to-peak,and the traditional first-order analysis such as noise signal probability,a soft decision mechanism of speech/noise is formed,which can effectively identify the speech excitation area in noisy speech signal.Computer simulation results indicate that when the input background noise is Gaussian noise and Gaussian like noise,the proposed algorithm has a lower probability of misjudgment than the traditional algorithm,and has better recognition performance under the condition of low signal-to-noise ratio.

关 键 词：LPC残差 高阶统计分析量 峰度 偏度 偏峰比 

分 类 号：TN912.34[电子电信—通信与信息系统]