Measurement and Prediction of Speech Intelligibility in a Virtual Chat Room

Intelligibility in a telecommunication-based conference application can largely be improved, when the speech signals of the different interlocutors are presented spatially, and when the sources are spatially distributed. For the controlled exploitation of the underlying Cocktail Party Effect in speech chat applications, a prediction of the intelligibility related to a given spatial configuration is sought. To this aim, a number of intelligibility tests have been conducted on different configurations of a virtual speech chat room based on binaural rendering technology and headphone presentation. The measurement object of the intelligibility tests was the Speech Reception Threshold, i.e. the signal-to-interferer ratio threshold for 50% intelligibility. Based on the test results, the paper presents a parametric model developed for intelligibility predictions in virtual chat-room applications, and briefly discuss its prediction performance.