Adapting television soundtracks for listeners with hearing loss

Abstract

This thesis investigates perceptually informed methods for adapting television soundtracks to improve accessibility for listeners with hearing loss. Conventional dynamic range compression, while widely used to restore audibility, often introduces artefacts such as pumping, distortion, and loss of natural timbre. The project explores alternative approaches, guided by auditory modelling, aiming to restore audibility while minimising perceptual side effects.

A framework was developed comprising an analysis stage, adaptation stage, and resynthesis. Input audio was decomposed using a complex gammatone filterbank, from which amplitude envelopes were derived. Envelopes were compared against both normal hearing thresholds and the thresholds of a model hearing loss to guide the amplification, local in both time and frequency. The implementation of smooth spectrotemporal amplification was achieved using a novel method, the Slow Inverse Fourier Transform, enabling reconstruction of time-domain signals after adaptation.

To gauge the extent to which quieter components of TV soundtracks needed their audibility to be preserved, a binary masking analysis was performed, removing elements deemed less perceptually significant according to a phons-based threshold. Results demonstrated that some low-level components inaudible to listeners with normal hearing could be selectively suppressed without compromising intelligibility, thereby reducing the need for global amplification when adapting for listeners with hearing loss.

This work demonstrates that perceptually informed processing can target audibility more precisely than conventional dynamic range compression, offering a pathway to improve comfort and intelligibility for listeners with hearing loss without compromising artistic intent.

Date of Award	Dec 2025
Original language	English
Awarding Institution	Queen's University Belfast
Sponsors	Northern Ireland Department for the Economy
Supervisor	Matthew Rodger (Supervisor) & Trevor Agus (Supervisor)