Dominant Voiced Speech Segregation and Noise Reduction Pre-processing Module for Hearing Aids and Speech Processing Applications

Speech experiences different acoustic obstructions in normal environment, whereas numerous of the applications require a compelling way to partitioned the original dominant speech from the impedance, a perfect hearing framework ought to be able to isolated and recognize sound-related occasions precisely from complex sound-related scenes and in unfavorable conditions. Difficulty in distinguishing a particular speech from a mixture of other unwanted conversations is one of the problems faced by people wearing hearing aid. The possibility of partition of overwhelming discourse from other discourse signals and its enhancement from that point will be accommodating for individuals with hearing disability. The recent literature in the Computational auditory scene analysis (CASA) systems are based on gammatone filter bank and Short time Fourier transform (STFT). But higher computational complexity associated with those models adversely affect the implementation of digital hearing aids. This paper introduces a cochlear model using Wavelet packet transform (WPT) and a novel approach for dominant voiced speech segregation. The experiments confirmed the enhancement of our model in terms of computational complexity and recognition rate when compared to competitive models.