Comparative study of speech enhancement algorithms and their effect on speech intelligibility

The aim of the Speech Enhancement (SE) algorithms is to recover clean speech from noise-corrupted speech. The applications of SE algorithms are in speech recognition, communication, signal compression and in the hearing aid devices. Speech Enhancement is done in several ways based on the type of noise, the parameter for enhancement, the domain of operation, the computational load. This paper discusses various speech enhancement algorithms consisting of spectral subtraction, multi-band spectral subtraction, statistical estimation, phase spectrum compensation, optimal filtering, discrete wavelet transform. The processed speeches from these SE algorithms are compared at the same set of conditions using five parameters like Perceptual Evaluation of Speech Quality, Short Time Objective Ineligibility measure, Minimum Mean Square Error, Itakura-Satio distance, Processing time. The noisy speeches are taken from Noizeus database with various noisy conditions and in different Signal to Noise Ratios.