Blind Separation of Acoustic Signals

This chapter presents an overview of criteria and algorithms for the blind separation of linearly mixed acoustic signals. Particular attention is paid to the underlying statistical formulations of various approaches to the convolutive blind signal separation task, and comparisons to other blind inverse problems are made. Several algorithms are described, including a novel algorithm that largely maintains the spectral content of the original mixtures in the extracted source signals. Numerical experiments are also provided to explore the behaviors of the algorithms in real-world blind signal separation tasks.