An algorithm has been developed which successfully separates a wanted voice from interfering voices and other sound sources. The system uses directional and harmonicity cues and is exceptionally robust to room reverberation. It does not rely on any prior information about the wanted or unwanted signals and is computationally efficient such that it is well suited to real time implementation in the future. A particular application is to provide an advanced hearing aid for hearing impaired, but the system could also be used as a pre-processor and enhancement unit for speech recognition and communication systems.