Compressed Sensing of Speech Signals Based on Linear Prediction Coefficients and Difference Transformation

On the research of compressed sensing, the sparse field by certain transformations is one of the most important factors on signal reconstruction. This paper presents a new united sparsity method based on Linear Prediction Coefficients (LPC) of speech signals, which associates LPC analysis with difference transform method. Orthogonal Matching Pursuit (OMP) algorithm is used to reconstruct the speech signal, and the reconstruction performance by this new method is compared with FFT and LPC. Experiments show that, when the compression ratio is larger than 0.4, the performance of reconstructed signal by united method is much better than the other two. Namely, when the reconstruction performance of the three methods is same, the compression ratio of the united method is less than that of the two, which means the united method has better compression performance. PESQ is used to evaluate the quality of reconstructed speech, and the speech reconstructed by the united method has the higher scores.