A Comparison of Computational Precedence Models for Source Separation in Reverberant Environments

Reverberation continues to be problematic in many areas of audio and speech processing, including source separation. The precedence effect is an important psychoacoustic tool utilized by humans to assist in localization by suppressing reflections arising from room boundaries. Numerous computational precedence models have been developed over the years and all suggest quite different strategies for handling reverberation. However, relatively little work has been done on incorporating precedence into source separation. This paper details a study comparing several computational precedence models and their impact on the performance of a baseline separation algorithm. The models are tested in a range of reverberant rooms and with a range of other mixture parameters. Large differences in the performance of the models are observed. The results show that a model based on interaural coherence and onset-based inhibition produce the greatest performance gain over the baseline algorithm. The results also show that it may be necessary to adapt the precedence model to the acoustic conditions of the room in order to optimize the performance of the separation algorithm.