PROCESSING OF SPEECH SIGNALS USING INDEPENDENT COMPONENT ANALYSIS

Independent Component Analysis (ICA) is a highly general statistical method in this paper in which observed random statistics are linearly translated into components that are maximally independent of each other and have "interesting" distributions concurrently. As an estimate of a latent variable model, the ICA can be formulated. The intuitive concept of maximal non-Gaussian can be used to extract various objective functions whose optimisation allows the ICA model to be estimated. Alternatively, more classical notions such as ultimate probability estimates or reciprocal knowledge should be used in estimating the ICA.