Applying Blind Source Separation and Deconvolution to Real-World Acoustic Environments

Sound engineers commonly use digital systems to record and analyze audio in music and film studios. Often, they need to cleanly access a single sound source such as an instrument or voice. While humans can focus their attention on any one sound source out of a mixture of many (a phenomenon termed in 1953 by E. Collin Cherry [1] as the “cocktail-party effect”), current digital audio systems lack this ability. This paper describes one way in which researchers in this field are approaching the goal of enabling a digital system to accomplish this task, which is commonly known as source separation. The main contribution of the work discussed in this paper is the application of current source separation algorithms to unconstrained real-world environments.