基于听觉感知的语音稀疏表示及压缩感知  

Sparse representation and compressed sensing for speech signal based on auditory perception

作　　者：陈砚圃[1] 张长青[1] 谭薇[1] 杨森斌[1] 

机构地区：[1]西安通信学院,西安710106

出　　处：《应用声学》2014年第1期53-59,共7页Journal of Applied Acoustics

基　　金：国家自然科学基金项目(61072125)

摘　　要：本文针对语音信号稀疏表示及压缩感知问题,将听觉感知引入稀疏系数筛选过程,用掩蔽阈值筛选重要系数,以得到更符合听觉感受的语音稀疏表示。通过对一帧浊音信号分别采用掩蔽阈值和能量阈值方法进行系数筛选对比实验,结果表明掩蔽阈值法具有更好的稀疏表示效果。为验证听觉感知对语音压缩感知性能的影响,与能量阈值法对照对测试语音进行压缩感知观测和重构,通过压缩比、信噪比、主观平均意见分等主客观指标评价其性能,结果表明,掩蔽阈值法可有效地提高压缩比且保证重构语音具有较高的主观听觉质量。This paper concerns the sparse representation and compressed sensing for speech signal, the auditory perception was brought into the selection of sparse coefficients to obtain a sparse representation which is more suitable to hearing. An experiment of sparse coefficients＇ selection under masking thresholds method was done comparing with the energy threshold method, the result showed the masking thresholds method was better. In order to validate the impact of the auditory perception model to compressed sensing for speech signal, the testing speeches were chosen to compress under compressed sensing framework by masking threshold method and energy threshold method, through the subjective and objective indicators, such as compression ratio, signal noise ratio and mean opinion score, a conclusion was made： the masking threshold method can lead a increasing of compression ratio while the quality of hearing for reconstructed signal is not decreased.

关 键 词：语音信号 稀疏表示 听觉感知 压缩感知 

分 类 号：TN912.3[电子电信—通信与信息系统]