Optimization of speech enhancement techniques coping with uncorrelated or correlated noises

Our concern is the enhancement of noisy speech for mobile radio communications when two observations are available. For a large distance between the microphones, the coherence between the noise is low while the coherence between the speech signals is close to one (in magnitude). Two methods have been proposed previously considering only uncorrelated noise. A modification is presented to take correlated noise into account. The aim is to study theoretically the methods to derive their attainable performance. Some results on real noisy speech are also given.