Adaptive Time Frequency Resolution for Blind Source Separation
Alexandra Crăciun, Martin Spiertz · RWTH Publications (RWTH Aachen) · 2010
In this article, we investigate the influence of adaptive time-frequency resolution schemes on a monaural blind source separation algorithm.The goal is to show that the capability of separating the original signals from the mixture is increased if we adapt the time-frequency resolution of the short-time Fourier transform to a certain mixture's characteristics.We will therefore implement different adaptive time-frequency schemes based on the usage of analysis windows of various lengths.The rules that define the mixing schemes rely on two measures: a transient detection measure by means of phase deviations and an energy concentration measure.Both algorithms are evaluated on a large test set, improvements being shown by use of objective quality measures.