| Summary: | The separation of independent sources from mixed observed data is a fundamental and challenging signal processing problem. In many practical situations, one or more desired signals need to be recovered from the mixtures only. A typical example is speech recordings made in an acoustic environment in the presence of background noise and/or competing speakers. Other examples include EEG signals, passive sonar applications and cross-talk in data communications. The audio signal separation problem is sometimes referred to as <I>The Cocktail Party Problem</I>. When several people in the same room are conversing at the same time, it is remarkable that a person is able to choose to concentrate on one of the speakers and listen to his or her speech flow unimpeded. This ability, usually referred to as the <I>binaural cocktail party effect</I>, results in part from binaural (two-eared) hearing. In contrast, a person with a severe hearing loss in one ear finds it is difficult to focus on a particular speaker under the same circumstances. A signal separation pre-process would be desirable in such circumstances. Signal separation techniques can also be applied in many other areas such as noise reduction, speech recognition and multi-media applications. The term 'Blind Signal Separation' refers to the lack of any propagation model: only statistical independence of the sources is assumed. The lack of other prior information underlines the difficulty of the problem. Observations may be modelled as linear mixtures of a number of source signals, i.e. a linear multi-input multi-output system. In this dissertation, the general n-source n-sensor (<I>n x n</I>) linear time invariant wide-band system is studied, in which, <I>n</I> random signals are received at <I>n</I> sensors and these signals originated from <I>n</I> sources. The problem is to recover the sources from observed signals only. Various block-based iterative algorithms are proposed which use output decorrelation as a signal separation criterion. These algorithms search for a linear transformation that minimises the statistical correlation between the components. Some existing solutions are reviewed and compared.
|
|---|
