Abstract
This work is part of the Perceptually Optimised Sound Zone project (posz.org) which aims to develop sound zoning systems which reproduce audio programmes to multiple listening zones within automotive and domestic environments. This work describes the construction of a model to evaluate sound zoning systems. A framework for evaluating auditory interference scenarios is described in which either the target or interferer programme is masked, or where both programmes are audible and the listening scenario has some degree of acceptability. Masking and acceptability experiments were conducted to investigate the relationship between the two, and to determine boundaries of audibility. A linear correlation was found between masking and acceptability, and a linear regression model was constructed to predict thresholds of acceptability from masking thresholds. A masking threshold model was adapted and predictions were within 3 dB of the reported mean masking thresholds. Predictions of acceptability, using a linear regression and masking model combination, accounted for three quarters of the variance in acceptability. Further work focused on speech target programmes based on listener comments that the presence of speech affected acceptability. An experiment was conducted to gather intelligibility and acceptability data. Results showed that a high speech intelligibility marked the lower boundary of acceptability. Existing models for intelligibility prediction were evaluated and a time-windowed speech intelligibility index was shown to predict intelligibility with RMSE = 10.8%. Subsequently, a model was constructed to predict acceptability within these boundaries. Two experiments were conducted gathering training and validation data, and a training and selection procedure was carried out to methodically identify the most useful features. The selected model predictions had acceptability scores of RMSE = 11.1-17.9% across training and validation data. Finally, an algorithm was proposed for the prediction of acceptability in auditory interference scenarios. The algorithm consists of first predicting masking thresholds to determine the boundaries of acceptability. Then, for non-speech target programmes, the acceptability is predicted using a linear regression to the masking threshold; for speech target programmes, the intelligibility is calculated to revise the lower acceptability boundary and the speech acceptability model is used to predict acceptability.