Describes various methods of data-driven control of hearing aid gain prescription as well as de-noising processing. Most of the work is on intelligibility prediction with an apriori refernce, but the last chapter descirbes an interesting technique which is "blind" to the reference speech and predicts intelligibility from the failure rate of a group of differently-trained speehc recognisers.
Period
23 Jun 2023
Examinee
Zehai Tu
Examination held at
The University of Sheffield
Degree of Recognition
National
Keywords
speech intelligibility, hearing aids, deep neural nets