Year: 2019
Type: article
Abstract: Normal-hearing subjects are able to localize and identify sound sources in reverberant multi-talker environments.It has been shown previously that subjects can accurately analyse reverberant multi-talker scenes with up to four simultaneous talkers.While multi-talker scene perception has mainly been investigated regarding only auditory information, visual information might influence the subjects' perception.In the present study, audio-visual scenes varying between two and ten talkers were considered.The acoustic information was provided using a spherical loudspeaker array and visual information was provided using head-tracked virtual reality glasses.The visual information represented 21 possible static talker locations and the subjects were asked to identify the content of the talkers and their specific locations.For the identification of talkers, the subjects were asked to label visual locations with headlines from the talkers' speech topics.The subjects were able to accurately analyse scenes containing up to six talkers.When more talkers were presented in the scene, the azimuth localization accuracy decreased, whereas distance perception accuracy was not found to vary with the number of talkers.This new audio-visual scene analysis method might be a valuable tool to test speech perception in more realistic environments than those tested in previous investigations.
Institution: Technical University of Denmark
Language: English
Topic: Speech and Audio Processing
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2996639980