Algorithms for direction specific Processing of Sound Signals - the Realisation of a binaural Cocktail-Party-Processor-System

Dissertation, Department of Electrical Engineering, Ruhr-University Bochum

Short title: Slatky, Direction selective Filters
For the Documentation: directional filtering - sound source localisation - binaural signal processing - Cocktail-Party-Processor - localisation - auditory model - psychoacoustics - speech signal processing - digital signal processing

This dissertation points out possibilities for a technical reproduction of the capabilities of human auditory system, to extract the sound signals of a certain direction ("one speaker") out of a mix of different sound sources ("Cocktail-Party").

Based on auditory experiments with several sound sources and following the signal processing methods of the human auditory system, Cocktail-Party-Processors have been developed, which determine, similar to the auditory system, levels and directions of active sound sources out of phase and level differences between two sound receivers. Interfaces have been prepared to include additional features of the auditory system, like the steering of the direction of attention in reverberant environment. ("Law of the first wave front").

With the help of these Cocktail-Party-Processors improvements of the signal-to-noise-ratio of up to 20 dB for signals of a desired direction can be achieved. Under free field conditions heavily disturbed speech signals, which are scarcely audible without processing, can be made understandable again.

Such a Cocktail-Party-Processor-System could be used, for example, as an input unit for speech recognition systems or hands-free telephones, as a noise suppression algorithm for hearing aids, as a noise source localisation and evaluation system.