摘要
非均匀临界频带与人耳对固有语音的频率感知特性十分相似,并且可在听觉感知上近乎完美地使信号进行重构。文章基于此提出了一种新的语音增强方法。该方法首先采用易于计算的无抽样离散小波包(UDWP)模拟人耳的听觉特性滤波器组对带噪语音进行分解;其次在语音信号的子带层次上用一种类似于软阈值的无穷阶可导的函数进行阈值处理;最后应用谱减法进行二次增强。实验结果表明,该方法在有效去除语音噪声的同时可以较大地提高语音的清晰度和可懂度。
Non-uniform critical bands,very much like human hearing system,can perform speech signal reconstruction almost faultlessly in auditory perception.The title of section 1 of the full paper is:implementing the filter set of UDWP(unsampled discrete wavelet packet).The core idea of such a set is that it uses parallel UDWP to decompose noise speech in almost the same way as human hearing system.The title of section 2 is:enhancing speech using UDWP.Its core idea is that,at the subband level of the speech signal,it...
出处
《西北工业大学学报》
EI
CAS
CSCD
北大核心
2009年第5期678-683,共6页
Journal of Northwestern Polytechnical University
基金
国家自然科学基金(10902084)
中国博士后基金(200902605)资助
关键词
无抽样离散小波包
语音增强
阈值
speech analysis
non-uniform critical bands
UDWP（unsampled discrete wavelet packet）
speech enhancement
threshold
