Generalized stochastic principle for microphone array speech enhancement and applications to car environments

In this paper we present novel solutions for microphone array speech enhancement systems that intelligently use the multipath environment to enhance signal coming from a desired location. We obtain a statistical principle that explains previously known factorization results of optimal beamformers, and proves a similar factorization holds for other new optimal estimators. Our solution requires a low computational load, and can be deployed on most of the platforms. We present speech recognition rates on real data, and compare a stereo versus a mono solution on this database.