基于小波变换和K-SVD的应急广播语音压缩方法  

Emergency broadcast speech compression based on wavelet transform and K-SVD

作　　者：鄢化彪[1] 胡超 黄绿娥 Yan Huabiao;Hu Chao;Huang Lyu’e(School of Science,Jiangxi University of Science&Technology,Ganzhou Jiangxi 341400,China;School of Electrical Engineering&Automation,Jiangxi University of Science&Technology,Ganzhou Jiangxi 341400,China)

机构地区：[1]江西理工大学理学院,江西赣州341400 [2]江西理工大学电气工程与自动化学院,江西赣州341400

出　　处：《计算机应用研究》2022年第11期3417-3421,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(11765008);江西省青年科学基金资助项目(20161BA216128);江西理工大学研究生创新计划资助项目(XY2021-S153)。

摘　　要：针对应急广播中语音传输效率低的问题,提出了一种基于小波变换和K-奇异值分解(K-SVD)的语音压缩方法,以提升应急广播信息传输的时效性。首先,该方法舍弃语音小波分解得到的高频分量,在小波合成时用随机信号代替;其次,在低频分量的压缩感知过程中,用K-SVD字典学习算法训练的过完备字典对其稀疏表示;最后,采用改进的基于子空间回溯的广义正交匹配追踪算法重构信号。实验结果表明,在压缩效率为50%时,该方法重构应急广播语音的客观语音质量评分(PESQ)达到3.717,比其他对照算法分别提升了3%~47%,说明在保证压缩效率的同时,所提出的方法能提升应急广播语音重构质量,确保应急广播的传输时效性。To address the problem of low efficiency of speech transmission in emergency broadcast,this paper proposed a speech compression method based on the wavelet transform and K-singular value decomposition(K-SVD) to improve the information transmission timeliness of emergency broadcast.Firstly,this method discarded the high-frequency components which was obtained from the speech wavelet decomposition,and replaced them with random signals during the wavelet synthesis process.Next,in the compressed sensing process of the low-frequency component,it was sparsely represented by an overcomplete dictionary which was trained by the K-SVD learning algorithm.Finally,this method reconstructed the signal by an improved algorithm of the generalized orthogonal matching pursuit based on subspace backtracking.Experimental results show that the perceptual evaluation of speech quality(PESQ) of the reconstructed emergency broadcast speech reaches 3.717 at a compression efficiency of 50%,the PESQ is 3% to 47% better than the comparison algorithms.It indicates that the method can improve the speech reconstruction quality of emergency broadcast and ensure the timeliness of emergency broadcast transmission while ensuring the compression efficiency.

关 键 词：语音压缩 压缩感知 K-奇异值分解 广义正交匹配追踪 

分 类 号：TN912.3[电子电信—通信与信息系统]