Hearing loss
Hearing
Dr Chris Scholes is exploring a new model for helping those with hearing loss identify voices in noisy environments - by using vocal tract movement.
One in five people who should wear hearing aids do not, and understanding speech in noisy situations is a common complaint.
Until recently, efforts at separating voice from noise focused on sound. However, it is well established that we understand speech better when we can see the speaker.
Seeing the speaker offers a source of information that could help to separate the voice from noise, improving hearing aid performance, and making hearing aids more attractive to the hearing impaired.
This team recently showed that vocal tract movement can be recovered from videos of speakers; the first step in being able to use 'facial speech' to improve hearing aids.
The team will collect a comprehensive speech database to test their model - recording the face, vocal tract and voice at the same time during speech.
The outcome will be a model that can predict vocal tract movement given either the voice or face or both. The team will also make our unique multimodal database available online for other scientists to explore.