基于自适应多尺度压缩感知的语音压缩与重构  

Speech Compression and Reconstruction Based on Adaptive Multiscale Compressed Sensing Theory

作　　者：孙林慧[1] 杨震[1] 叶蕾[1] 

机构地区：[1]南京邮电大学通信与信息工程学院,江苏南京210003

出　　处：《电子学报》2011年第1期40-45,共6页Acta Electronica Sinica

基　　金：国家863高技术研究发展计划(No.2006AA010102);国家自然科学基金(No.60971129No.60902065);南京邮电大学青蓝计划(No.NY208038);江苏省普通高校研究生科研创新计划(No.CX09B-148Z;No.CX10B-191Z;No.CX10B-189Z)

摘　　要：本文针对语音信号的压缩感知问题,在系数总长度不超过原信号长度的前提下,推导了Sym小波分解合成的矩阵形式,提出了语音信号多尺度压缩感知(MCS)框架.进一步分析语音信号在小波基下不同级的稀疏性,提出了自适应多尺度压缩感知(AMCS)方法,把该方法运用到语音压缩与重构中,对重构语音进行了主客观评价,并进行了说话人识别验证,得出结论:基于AMCS比三层MCS重构语音的性能好.In this paper,a matrix form of Sym wavelet decomposition and synthesis is deduced,keeping the length of the coefficient no more than the length of original speech signals,and then we propose a framework of speech Multiscale Compressed Sensing（MCS） and an Adaptive Multiscale Compressed Sensing（AMCS） method by analyzing sparsity of different wavelet levels of speech signals.We compare AMCS with MCS by applying both methods to speech compression and reconstruction,and the reconstructed speech signal evaluated by the objective and subjective evaluation is applied to speaker recognition.The experimental results show that the reconstruction performance of speech signal based on AMCS is superior to MCS.

关 键 词：Sym小波 多尺度压缩感知 自适应多尺度压缩感知 语音压缩与重构 基追踪 

分 类 号：TN912.3[电子电信—通信与信息系统]