A Bandpass Filter With Multi Deep Denoising Autoencoder for Hearing Applications

Abstract

Speech enhancement techniques in hearing applications aimed to improve the quality of speech in a noisy environment. Deep denoising autoencoder suppresses noise from noise corrupted speech efficiently. Unfortunately, previous applications provide only limited benefits for the enhancement of speech in noisy environments. This paper presents a new approach for the hearing application, which indicates two stages of the bandpass filter and a model composed of three levels of deep denoising autoencoders. In the first stage, the bandpass filter designed to allow signals based on the human cochlea, which then followed by a model of three levels of multilayers deep denoising autoencoder, each which specialized for specific enhancement task of a complete set of tasks. The approach performance measured using the perceptual evaluation of speech quality, hearing aid sound quality index, and segmental signal-to-noise ratio. The simulation results prove that the proposed method yielded higher intelligibility and quality in comparison with single-multilayers neural networks.