Dynamic multiband filter Structures for simultaneous Improvement of speech quality And intelligibility

Abstract

Speech often occurs in a noisy environment and can thus be newlinedegraded by noise either during its production farend speech or its perception newlinenearend speech The resultant speech exhibits a deterioration in its significant newlineperceptual attributes namely quality and intelligibility Speech quality refers to newlinethe naturalness and cleanliness of speech whereas speech intelligibility denotes newlinethe underlying contents of the utterance In order to improve these attributes newlinespeech enhancement algorithms that involve removing the noise components in newlinethe corrupted speech are employed However the existing speech enhancement newlinealgorithms either improve the quality of the degraded speech or its intelligibility newlinebut not both simultaneously Owing to the uncorrelated nature of quality newlineand intelligibility quality improvement achieved by the conventional speech newlineenhancement algorithms does not guarantee an improvement in intelligibility. newlineThe current work focusses on the simultaneous improvement of quality newlineand intelligibility with minimised amount of deterioration in them by newlinedefining multi-band filter structures based on speech noise and perceptual newlinecharacteristics which are dynamic to accommodate the timevarying nature of newlinespeech and noise newline newline