Overview

Sensorineural loss results in a reduced dynamic range of hearing and widening of auditory filters resulting in degraded speech perception particularly in noisy environments. Hearing aids with dynamic range compression are used to compensate for reduced dynamic range. Hearing aid fitting, involving the setting of user-specific compression parameters, is generally carried out in environments with low ambient noise. These hearing aids improve speech perception in quieter backgrounds. However, listeners with hearing impairment do not find the performance of these aids satisfactory in day-to-day noisy situations. Speech perception in presence of background noise can be improved by incorporating speech enhancement techniques into hearing aids. Some of these techniques that can be explored are listed below.

Single-channel speech enhancement to improve speech quality and intelligibility:

Here, the objective is to develop a time-domain machine learning based speech enhancement technique that results in magnitude and phase enhancement.

Noise aware speech enhancement:

It is known that deep learning based speech enhancement techniques perform well on data that has similar characteristics to the data used for training. However, these techniques do not perform satisfactorily when the test data is unseen or has different distribution than that of the training data. The aim is to develop noise aware speech enhancement technique to improve the performance on unseen data. Also, techniques for incremental learning will be explored to improve the performance when the noise data distribution changes with time.

Multichannel Enhancement:

Hearing aids generally use multiple microphones that are suitable for a multichannel beamforming-based speech enhancement solution. Beamforming techniques perform spatial filtering and aim at separating the target speech from noise source located in a spatially different direction. The aim here is to develop a beamforming technique that does not require knowledge of reference microphone needs for use in hearing aids.

Real-time implementation and evaluation:

The speech enhancement techniques need to be implemented for real-time processing for use and evaluation by listeners with hearing impairment. Development of the proposed techniques on mobile and embedded devices such as on a smartphone will provide an affordable alternative to hearing aids and result in faster evaluation by users.