Chapter 19 - Audio applications

Abstract

This chapter discusses the application of the blind source separation (BSS) techniques to the separation of audio signals, with main emphasis on convolutive independent component analysis (ICA) and sparse component analysis (SCA). The need for BSS arises with various real-world signals, including meeting recordings, hearing aid signals, music CDs, and radio broadcasts. These signals are obtained via different techniques, which result in different signal properties. In all source mixing situations, the objective of BSS is to extract one or several source signals from the observed multichannel mixture signal, with other source signals being regarded as undesired noise. The signals of interest depend on the application, for instance, in the context of speech enhancement for mobile phones, the only source signal of interest is the user's speech. Undesired sources may then include speech signals from surrounding people and environmental noises produced by cars, wind, or rain. On the contrary, the so-called cocktail-party application refers to the situation when the observed mixture signal results from several people simultaneously speaking in a room and all speech signals are of interest. Noise may then originate from clinking glasses or footsteps.