Output details

Adaptive Sparsity Non-negative Matrix Factorization for Single-Channel Source Separation

This research successfully separates speech mixtures recorded from a single microphone. The algorithm has been adapted to analyse data in a multidisciplinary project MRC project ‘Monitoring Devices in Old Age Depression’ (G1001828 £247,580) with the Newcastle medical school. Complex structures of natural signals require adaptive tools to make use of their intricate redundancies. Researchers advocate using sparsity and overcomplete signal/image representations. However these generic methods have limited computations efficiency or theoretical ability to extract specific patterns. This paper is significant as it uses a variational Bayesian approach to compute the sparsity parameters for optimising the matrix factorization.