RETRACTED ARTICLE: Deep convolutional neural network-based speech enhancement to improve speech intelligibility and quality for hearing-impaired listeners

Abstract

In this paper, we propose a deep learning-based speech enhancement (DLSE) method to improve speech intelligibility for the hearing-impaired listeners. The algorithm decomposes the noisy speech signal into frames (as features) and feeds them to the deep convolutional neural networks (DCNNs) to produce an estimation of which frequency channels contain more perceptually important information (higher signal-to-noise ratio, SNR). This estimate is used to attenuate noise-dominated and retain speech-dominated cochlear implant (CI) channels for electrical stimulation, as in traditional n-of-m CI coding strategies. The proposed algorithm was evaluated by measuring the speech-in-noise performance of 12 CI users using two types of background noises such as fan and music sounds. The architecture and low processing delay of the DLSE algorithm make it suitable for application in hearing devices. While DLSE was evaluated using a noise-specific approach, several aspects of generalisation to unseen acoustic conditions were addressed, most importantly performance with a speaker not used during the training stage. The largest improvements for both speech intelligibility and quality are found by DCNN-based proposed method. Moreover, the results show that DCNN-based methods appeared more promising than existing methods. [Figure not available: see fulltext.]. 2018, International Federation for Medical and Biological Engineering.