A listener preference model for spatial sound reproduction, incorporating affective response.

Abstract

This work presents results and models for listener preference to music delivered via different spatial reproduction formats, here via mono, stereo and multichannel (5.1-ch) reproduction. Although this problem has been researched in the past, the current work introduces an elaborate multistage experimental procedure which considers the contribution of listener-specific emotional responses (valence and arousal) to his/hers Overall Listening Experience (OLE). The test procedure registers the individual listener preference and familiarization with the content of each test audio sample. A spatial envelopment metric, extracted directly from each audio signal sample is utilized as attribute for the perceived differences between the 3 different systems. This attribute, along with listener content preference (for each music sample) and his/hers affective response attributes are combined into linear regression model(s) which can predict the dominant trends for rating OLE. A novel linear tree approach is also proposed which highlights additional associations between the attributes within this multidimensional space. Comparative performance analysis shows that the proposed linear tree approach also achieves improved predictions for OLE ratings.