Abstract: Two or more microphones are mounted in an environment that contains an equal or lesser number of distinct sound sources. Acoustic energy from each source, with its attendant echoes and reverberation, impinges on each microphone. Using direction-of-arrival information, a first module attempts to extract the original source signals as if the acoustic environment were anechoic. Any residual crosstalk between the channels, which may be caused by echoes and reverberation, is removed by a second module. The first and second modules may be implemented using existing technology.

Abstract: This invention is a component of an adaptive filter signal processor whose purpose is to separate a mixture of signals received by each of a plurality of transducers. This invention accomplishes a key step in that separation: it estimates, from the transducer signals, the relative propagation delays among the transducers for each source. First, it randomly generates a fixed number of sets of delay parameters, called a population. Thereafter, the following events occur at each clock cycle. Each set is processed by an instantaneous-performance calculator, to generate an instantaneous-performance value. The instantaneous performance value of each set is added to a cumulative performance value. The set with the greatest cumulative performance value is transferred to the adaptive filter signal processor. New parameter sets are generated at random.