基于C1C2复杂性非线性特征的语音端点检测  

Application of C1C2 Complexity Measure in Detecting Speech

作　　者：庞全[1] 吴秀良[1] 杨翠容[1] 

机构地区：[1]杭州电子科技大学生物医学工程与仪器研究所,浙江杭州310018

出　　处：《电声技术》2008年第8期60-63,共4页Audio Engineering

基　　金：国家自然科学基金(60302027)

摘　　要：研究基于C1C2复杂度非线性特征的语音端点检测技术。应用中心削波法对含噪语音信号进行预处理,分析状态空间分割方法和窗长对检测性能的影响;通过允许字和禁止字来度量语音序列,采用C1C2复杂度对含不同噪声类型及不同信噪比的中英文语音样本进行了端点检测实验。在低信噪比情形下,检测方法要优于C0复杂度特征检测方法,方法具有较优的稳健性和实时性等特点,为强背景噪声下的语音端点检测提供了新的研究途径。A new speech detection method based on C1C2 complexity that has nonlinear properties is proposed. The noisy speech is preprocessed using the center clipping method. The influence of state space partition method and window size on detecting performance is analyzed. The comparison experiments of speech signals corresponding to different SNR and noise type are designed through measuring the complexity behaviors on distinct excluded blocks and distinct included blocks. Simulating results indicate that the method has better performance than CO complexity especially in detecting low-SNR signal. This method has good robustness and strong real-time capabilities. A new approach for detecting endpoints of weak signals is given under noisy environment.

关 键 词：语音端点检测 C1C2复杂性 状态空间分割 

分 类 号：TN912[电子电信—通信与信息系统]