There are many signals produced by the human body which carry important information of its health and emotional, state, intentions, will and several other aspects. These signals are of nonlinear and nonstationary nature, and the information is hidden and coded in a complex way which is not clearly exposed in the raw signal. Those properties degrade the performance of automatic analysis and classification systems using those signals. We use recent advanced signal processing methods like sparse representation, nonnegative matrix factorization, blind source separation, independent component analysis, regularization and inverse problems to enhance the capabilities of information extraction and improve the behavior of systems for computational analysis of that information. The objective of this research line is to produce advanced signal processing and machine learning algorithms capable of extract that information, for a wide range of applications in speech signals, in respiratory, cardiac and electroencephalography signals, among others.
fdICA: Algorithm for frequency-domain blind source separation based on the pseudoanechoic model.
multibinICA: Algorithm for frequency-domain blind source separation based on Multibin ICA for 2 by 2 mixtures.
blindder: Semi-supervised method for microRNA prediction with few labelled samples.
Beta Dereverberation: A blind, single channel dereverberation method in the time-frequency domain.