Speech Signal Compression and Reconstruction Using Compressive Sensing Approach

Abstract

There are various data compression techniques available in the literature. But we have to follow Shannon’s theorem if we want efficient recovery of the desired signal after compression. If we follow Shanons’s theorem, then the rate at which we perform the sampling must be more than double the highest frequency component in the signal. But as we know that there are various applications present nowadays in which we have to store or transmit lots of data like MRI, RADAR, satellite communication, and other data transmission process, because of that this is so much tedious to follow Shannon’s theorem. To overcome this problem, researchers came up with a new approach which is known as compressive sensing, and if a signal is sparse in any domain, then a compressive sensing approach can be easily applicable. According to compressive sensing theory by using very few nonzero value samples, we can easily reconstruct the original signal. In this paper, the same compressive sensing approach is applied to the ten different words that are said by a single speaker. Basis pursuit algorithm is applied to reconstruct the original speech signal. To estimate the quality of speech signal, two approaches are used in this paper, one is signal-to-noise ratio (SNR) and another is mean square error (MSE).