基于语音特征的自适应压缩感知  

Adaptive Compressed Sensing Based on Speech Characteristics

作　　者：刘智威[1] 吴晓欢[1] 朱卫平[1] 

机构地区：[1]南京邮电大学信号处理与传输研究院,江苏南京210003

出　　处：《南京邮电大学学报（自然科学版）》2013年第1期39-43,61,共6页Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition

基　　金：国家自然科学基金(60972041);江苏省普通高校研究生科研创新计划(CXZZ11_0397)资助项目

摘　　要：利用语音在DCT域的稀疏性,提出了一种基于语音分为清音和浊音的特点,自适应分配观测点数的语音重构方法。首先根据清浊音在整个语音段占有的能量比分配观测点,然后判断每帧语音性质。如果是清音,则根据能零比的大小来分配该帧的观测点数;如果是浊音,则根据能量的大小来分配观测点数。实验表明:语音信号是稀疏的并且可压缩,在同种压缩比下,文中所采用的语音重构算法具有较好的信噪比、误差以及MOS分。Utilizing the sparsity of the speech signals in the DCT domain and considering the composition of voiced and unvoiced speech, a speech recovery method is proposed with adaptively distributed number of measurement points. It distributes the measurement points according to the energy ratio of voiced speech and unvoiced speech and then judges on the property of each speech frame. If the frame is unvoiced speech, the number of measurement points is calculated according to its zeros and energy rate. If the frame is voiced speech, the number of measurement points is calculated according to its energy. The experiment results demonstrate that the recovered speech signal resulting from the proposed method has good signal to noise ratio, high MOS score and low error.

关 键 词：压缩感知 语音重构 稀疏性 OMP 自适应 

分 类 号：TN912.3[电子电信—通信与信息系统]