Modeling the separation of acoustical signals based on the simulation of spike firing patterns
Kyrill A. Fischer · The Journal of the Acoustical Society of America · 1994
It is investigated to what extent the separation of overlapping acoustical sounds may be achieved by detecting and grouping comodulated frequency bands. Comodulation is defined by coherent envelope fluctuations which are detected by means of physiological motivated algorithms involving spike representation and coincidence detection. Groups of comodulated frequency bands are defined by cluster analysis and their temporal variations are traced. Every group detected can then be separated from its ‘‘acoustical background’’ by suppressing all channels not belonging to the given group. The paper addresses problems concerning the definition of suppression gain values and the problem of overlapping groups.