帧间自适应语音信号压缩感知

摘要

Compressed Sensing (CS) is a recently proposed theory that enables the exact reconstruction of signal sampled via sub-Nyquist sampling rate. The theory has been applying for simplifying the traditional sampling hardware, reducing sampling time consumption and decreasing storage space of data. Benefiting from the superiority of CS technique for speech signal transmission, we propose an adaptive inter-frame speech compressed sensing method. Under the assumption that speech signal is sparse in Discrete Cosine Transform (DCT) domain and according to the statistical behavior of speech signal frames, our proposed method takes into account both intra-frame energy and inter-frame consecution of location in our a-daptive compressed sensing algorithm. Experimental results show that, the method of intra-frame energy adaption can promote the speech recovery quality apparently and the method of location adaption can reduce the speech recovery time obviously. Namely, the proposed adaptive compressed sensing algorithm in this letter can achieve higher speech reconstruction performance with lesa time consuming.%近年来提出的压缩感知是一种以低于传统奈奎斯特速率对信号采样可得到精确恢复的理论.该理论很快应用于简化传统的采样硬件、缩短采样时间、以及减少数据的存储空间.针对语音信号的传输问题,本文提出一种帧间自适应语音信号压缩感知的方法.在离散余弦变换域的语音信号具有稀疏性的前提下,以大量语音信号帧的分析统计为依据,提出一种基于语音帧能量分级和帧间位置惯性的语音信号自适应压缩感知算法.实验结果表明,能量自适应可以显著地提高语音信号的恢复质量,而位置自适应可以明显地减少语音信号的恢复时间,从而本文提出的算法可以用较少的恢复时间获得较好的恢复效果.