Supervision: Qin Liu

Project type: Master thesis

Available

The ability to perceive sounds in space — a concept known as spatial hearing — is a function of our auditory system that relies on the utilization of several acoustic cues. In particular, the binaural structure of the auditory system (two ears) allows for the comparison of the sound reaching each eardrum to estimate sound direction. In the field of hearing aids, the quality of spatial hearing is crucial for accurately localizing sound sources. For this reason, the design of various algorithms for hearing aids should consider the preservation of auditory spatial perception cues in order to provide the best experience to the listener. Some hearing aid processing might indeed distort some of the important cues necessary for accurate sound localization.

Traditionally, assessing the effect of new algorithms on the spatial perception of hearing aids is typically achieved by performing subjective listening experiments. Conducting such tests is costly, time-consuming, and requires a long process: obtaining ethical agreement, recruiting participants, running individual listening experiments, etc. Thus, it is desirable to investigate the possibility of using objective metrics obtained with spatial perception models to evaluate the preservation and degradation of important cues for auditory localization. Essentially, this project is about constructing a toolbox to simulate human auditory localization behaviour. This will equip us with a practical tool to assess the performance of hearing aid algorithms, thus eliminating the need for conducting complex listening experiments.

The project consists of three stages. The first stage involves building a toolbox for 3D sound localization which considers the impact of hearing loss. The second stage will involve measuring the Head Related Transfer Function (HRTF) of a dummy head (Kemar) to test the proposed algorithm with people who have normal hearing. The third stage will conduct a listening test to validate the toolbox on a group of approximately 10 individuals with normal hearing and 20 people with hearing impairment. At this stage, the emulation of hearing loss behaviour on sound localization will be evaluated.

Profile: Electrical engineering, Physics

Prerequisites: Acoustics, Signal Processing

Learning outcome: Acoustic signal analysis, machine Learning algorithms, auditory measurement

Context: Coding (30%), Experiments (40%), Listening tests (30%)