Electronic Thesis and Dissertation Repository

Thesis Format

Integrated Article

Degree

Master of Engineering Science

Program

Electrical and Computer Engineering

Supervisor

Parsa, Vijay

Abstract

Hearing aids use a variety of noise reduction techniques to enhance the experience of hearing impaired listeners. One of these techniques is beamforming, which typically aims to preserve sounds coming from the front of the user and suppresses those from the sides and back. Recently, hearing aids have begun employing a wireless connection between the left and right hearing aids in order to augment the directionality of the beamformers, called binaural beamformers. However, the effect of these binaural beamformers on perceived quality and intelligibility has not been thoroughly tested. This thesis investigated the benchmarking of hearing aids which utilize binaural beamforming algorithms using behavioural testing and computational models. Speech recordings from bilateral pairs of several popular hearing aids were obtained across different processing conditions, and in different noisy and reverberant environments. The quality of these recordings was evaluated subjectively by thirteen hearing impaired adults. In addition, computational predictors of perceived quality and intelligibility were extracted from the left and right hearing aid recordings. Objective and subjective analyses revealed that binaural beamforming has a generally positive effect on quality and intelligibility that was dependent on the directionality of the speech and noise. The ear recording with the better predicted quality score was also found to correlate better with the subjective quality ratings than the average of left and right ear predicted scores. A new weighting function that optimally combines the monaural computational metrics was developed, which was shown to be especially effective in environments where speech and/or noise sources are asymmetrically positioned.

Summary for Lay Audience

Hearing aids use a variety of signal processing techniques to enhance the experience of hearing-impaired listeners across varied listening environments. One of these techniques is noise reduction, where unwanted signals such as ambient noise and unwanted speech are suppressed while signals such as wanted speech are enhanced. Recently, hearing aids have begun utilizing binaural beamformers, which use a wireless link between the left and right hearing aids in order to amplify signals originating from the front of the user while suppressing those from the sides and back. Effectively, the algorithm utilizes the assumption that the user is looking at what they want to listen to in order to reduce noise. However as binaural beamformers have only been recently developed, the actual benefit the algorithms have on enhancing the quality and intelligibility of speech in noisy conditions is largely unknown. This thesis investigated the benchmarking of hearing aids which utilize binaural beamforming algorithms using both computational models of the auditory system as well as behavioural testing with hearing-impaired listeners. Binaural beamformers were found to have a generally positive effect on the quality and intelligibility of speech, however it largely depended on the directionality of the speech and noise. It was also found that when using a computational model to predict speech quality, the better scoring ear was a better predictor of the behavioural testing results. A new weighting function to combine predicted quality scores for the left and right ears was developed that more heavily weights the better scoring ear.

Available for download on Friday, August 21, 2020

Share

COinS