Real acoustic environments are often the result of a complex mixture of different audio sources. In many speech and audio applications, only one of thesources is the signal of interest, while all remaining components act as a noise source, usually degrading the performance of the processing system. This scenario includes the well known "cocktail party" problem.
Algorithms for separation and identification of individual sources or reduction of interfering signals have been a goal of audio processing systems since the early 60s. Recently, the concept of Independent Component Analysis (ICA) and properties as the sparsity of speech in the frequency domain provided some new approaches to the problem of blind separation of acoustic mixtures. In this presentation we will make a brief overview of existing audio separation techniques, starting with conventional methods and reviewing the main contributions of ICA methods and other recent techniques to the field.