量化压缩感知在语音压缩编码中的应用

摘要

利用语音信号在离散余弦变换( DCT)域的近似稀疏性和量化压缩感知理论,文中提出一种基于量化压缩感知的语音压缩编码方案。编码端利用压缩感知技术,将语音信号投影成数据量大大减少的观测序列,然后对观测序列采用Lloyd-Max量化得到量化后的观测样值；解码端直接利用量化后的观测样值,结合重构算法重构出原始语音信号的DCT系数,经过DCT反变换得到重构后的语音信号,并采用后置低通滤波器改善重构语音的听觉效果。该编码方案解码端不需要进行反量化,而是直接利用量化后的观测样值进行重构,有效降低了解码端的运算量及复杂度。仿真结果表明：采用量化迭代硬阈值(QIHT)算法重构效果优于迭代硬阈值算法(IHT),重构语音的信噪比能达到20 dB以上,MOS分达到3.26。%Utilizing the sparsity of Discrete Cosine Transform ( DCT) coefficients of speech signal and the theory of quantized compressed sensing,a novel speech coding scheme based on Quantized Compressed Sensing ( QCS) is proposed in this paper. Based on CS theory, the speech signal is transformed into measurement sequence at encoder side,by which the size of the data set is significantly reduced. Af-ter quantizing the measurement sequence by Lloyd-Max scheme,the sample value of measurement is gained. At decoder side,together with reconstruction algorithm,the DCT coefficients can be reconstructed by the measurements. The speech signal can be reconstructed af-ter DCT inverse transform. The quality of reconstructed speech signal can be improved by post low-pass filter. The speech signal can be directly reconstructed by the quantized measurements without inverse quantization,leading to the reduction of computation and complexity in decoder site. The simulation results show that the performance of Quantized Iterative Hard Thresholding ( QIHT) algorithm is superior to that of Iterative Hard Thresholding ( IHT) algorithm. The Signal-to-Noise Ratio ( SNR) of the reconstructed speech signal is about 20 dB while mean opinion score (MOS) is up to 3. 26.