Modelling binaural processes involved in simultaneous reflection masking: limitations of current models

Masked thresholds were measured for a single test reflection, masked by the direct sound, as a function of the reflection delay. This was done for diotic as well as for dichotic stimulus presentations and all stimuli were presented via headphones. The input signal was a 200-ms long broadband noise. The dichotic threshold decreased with increasing reflection delay indicating an increase in binaural detection performance with increasing reflection delay.

Comparing the dichotic threshold to the corresponding diotic threshold, for delays below 7-10 ms, the dichotic threshold was found to be higher than the diotic threshold while it was lower than the diotic threshold for larger delays. Hence, the binaural system seems to deteriorate auditory detection performance for very early reflections and to enhance auditory detection performance for later reflections.

Existing binaural (detection) models, such as normalized cross-correlation models (e.g., Bernstein et al., 1999, JASA, pp. 870-876), the power-addition model (Zurek, 1979, JASA, pp. 1750-1757), or Equalization-Cancellation-based models (e.g., Breebaart et al., 2001, JASA, pp. 1074-1088), cannot account for the psychoacoustical data.

The present talk aims at understanding why these binaural models in their current form cannot describe the binaural mechanisms involved in reflection masking and a number of model-modifications are discussed that might help to overcome this deficiency.