M2-ESR: A computational model of neuronal responses for speech enhancement

Inspired by results from recordings in the mammalian auditory brainstem we suggest a novel speech enhancement strategy. Neural adaptation can be described in a simplified form based on fractional derivatives of the sound envelope. In this project we aim to develop a transformation into the ‘neural space’ as an approach to overcome the restriction of classical algorithms which concentrate on energy. We have demonstrated already that a supervised sparse coding scheme works in the neural space and that the sound can be resynthesized with reduced noise and intact quality.

Host institution: University of Southampton (SOTON)
Supervisor names: Stefan Bleeck and Steve Eliott
Start date: 1.6.2013
PhD enrolment: Yes
Employment contract: Full-time early-stage research fellow, 36 months