Abstract
Audio-on-audio interference situations are a common occurrence in everyday life; they may be naturally occurring or be a side-effect of a non-ideal personal sound zone system. In order to evaluate and optimize such situations in a perceptually relevant manner, it is desirable to develop a model of listener experience. Distraction ratings were collected for 100 randomly created audio-on-audio interference situations with music target and interferer programs. A large set of features was also extracted from the audio; the feature extraction was motivated by a qualitative analysis of subject responses. An iterative linear regression procedure was used to develop a predictive model. The selected features were related to the overall loudness, loudness ratio, perceptual evaluation of audio source separation (PEASS) toolbox interference-related perceptual score, and frequency content of the interferer. The model was found to predict accurately for the training and validation data sets (RMSE of approximately 10%), with the exception of a small number of outlying stimuli.