The general problem category addressed by our Phase I SBIR project concerned the so-called cocktail party problem, which has long been recognized as an important but difficult speech processing/understanding problem. It arises when we wish to process single-microphone (monaural) or multi-microphone (multi-aural) data collected in a reverberant environment in order to separate a speaker’s voice from other voices and sounds in the background. Obviously, the human brain is able to perform such processing even when the background voices and sounds are quite loud, but the important question is whether a machine can be endowed with a similar
capability.
Copy of SBIR Results of the Phase I
[Note: A more detailed version of Phase I results is given in our Phase I Final Report].