Nonlinear perception of hearing-impaired people using preference learning with Gaussian Processes
Year: 2008
Type: article
Abstract: In this report, we describe a probabilistic kernel approach to pairwise preference learning based on Gaussian Processes proposed in Chu and Ghahramani [2005] and apply the method to audio input signals. The objective is to have a principled approach for modeling and pre-dicting speech quality for arbitrary degradation mechanisms that might be present in a hearing aid. In Arehart et al. [2007] pairwise comparisons were performed with 14 normal-hearing and 18 hearing-impaired subjects for several sound distortions. The kernel approach gives a sig-nificant improvement in the prediction of sound quality perception over previous results. We show a significant difference between normal-hearing and hearing-impaired subjects, because of nonlinearities in the perception of hearing-impaired subjects. In this report, we describe a probabilistic kernel approach to pairwise preference learning based on Gaussian Processes proposed in Chu and Ghahramani [2005] and apply the method to audio input signals. The objective is to have a principled approach for modeling and predicting speech quality for arbitrary degradation mechanisms that might be present in a hearing aid. In Arehart et al. [2007] pairwise comparisons were performed with 14 normal-hearing and 18 hearing-impaired
Authors: P.C. Groot, Tom Heskes, Tjeerd M. H. Dijkstra
Institution: Radboud University Nijmegen
Language: English
Topic: Speech and Audio Processing
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2109877883