The cocktail party problem [instrumentation notes]

Consider a cocktail party with many people attending and many conversations going on simultaneously. If we recorded the sound from somewhere in the room, we would find that we had recorded many conversations at once, and it would be very difficult to tell what was being said by any given person. The problem of taking such a recording, or several such recordings, and separating out the different speakers is known as the cocktail party problem. The generalization of this problem to separating different independent components of a signal without making use of any specific knowledge of the component signals is known as blind source separation. The ability to take a recording in which many people are speaking and to create separate tracks that contain what was said by each person is of great interest to industry and government.