基于近似KLT域的语音信号压缩感知.pdfVIP

电子与信息学报 VoJ. 31No.12 第 31 卷第 12 期 2009 年 12 月 Journal of Electronics Information Technology Dec.2009 基于近1以 KLT 峨的谐音倍号压缩感知 郭海战 杨震 (南京邮电大学信号处理与传输研究院 南京 210ω3) 摘 要:ffi缩感知是近年来兴跑的研究然点，该文基于语音恼号在 KLT 域的稀疏特性，提出了3墨子模板肌配的近 似KLT，并在基于棋板阻自己近似 KLT 域上研究了语音信号的压缩感知性能。首先验证语音信号在J辈子模板匹配近 似KLT 城上的稀疏性，然后由语寄信号与观测矩阵构造相应的观测，采取回定分配每帧观测个数和按帧能最臼适 应分配每帧观恻个数两种方案，再以观测为已知条件利用L1优化算出重构语齿伯号在每豆子模扳肌配近似 KLT 域 的稀峭系数向囊，进而王震构原始谐音信号。实验表明，谐音倍号在基于模板匹配的近似KLT 域的压缩感知性能较 好。 3夜锦诩z 语音合成t 压缩感知:稀疏性; L1 优化; Karhunen阳Loeve 变换(KLT) 中阁分类号; TN912.3 X献标识码:A 文章编号; 1009-5896(2009)12-2948-05 Compressed Speech Signal Sensing ßased on Approximate KLT 。uo Hai-yan Yang Zhen (Institute 01 s机α1 Processing and Transmission, Nanjing University 01 Posts and Telecommuniωtions， Nα时ing 210003 , Chinα) Abstract: Comprωsed Sensing is a rωearch fOCU8 rising in recent years. On the basis of the signaJ s sparse representation in the KLT domain ,this paper propo响s an approximate KLT method using template matching and studies on the corresponding comprωsed speech signal sensing. First , it verifies the sparsity of sp朋ch signal in the approximate KLT domain. Second，问r speech signaJ and a measurem白nt matrix , it arranges measurements of fIXed or adaptive length according to fra.me energy. Third , according to the me皿urements， it finds the speech sign础吕 sparsest coefficient vector through L1 optimization algorithm to recover the speech signal. Simulation results demonstrate that compressed sp嗣ch signal sensing in the approximate KLT using template matching h制 good performance. K町 words: Sp制ch synthesis; Compressed Sensing (