Speech Enhancement in Non-stationary Noise Using Compressive Sensing

This paper addresses the problem of single channel speech enhancement algorithm in non-stationary noise environment which is rather difficult compared to the stationary noise. We proposed a new speech enhancement algorithm based on compressive sensing. First, the noise average estimation and Wiener filter gain are calculated. Compressive sensing using GPSR technique is then incorporated by randomly selected the sparse signal of unconstrained problem with suitable basis and reconstruct the noiseless distortion to the enhanced speech. The performance is evaluated using PESQ score improvement. Our proposed algorithm shows better performance compared to other traditional algorithms across two non-stationary noises at various SNRs. On average, the PESQ improvement was 19.14% and 7.12% for exhibition and restaurant noises, respectively.