Auxiliary-Function-Based Dynamic Independent Vector Analysis for Blind Extraction of a Single Moving Source

We extend independent vector analysis (IVA) to blindly extract a single moving source from its mixture with other spatially stationary sources. In this single-moving-source scenario, only the mixing vector corresponding to the moving source is time-varying, whereas the remaining mixing vectors are assumed to be time-invariant. We show that the inverse of the resulting mixing matrix (i.e., the time-varying separation matrix) can be reparameterized as the product of a time-invariant matrix and a matrix with only one time-varying column. This representation substantially reduces the number of time-varying parameters and enables tractable estimation of the full time-varying separation matrix. Furthermore, we show that the iterative source steering algorithm developed for auxiliary-function-based IVA can be naturally extended to this single-moving-source setting. Experimental results demonstrate that the proposed extension of IVA achieves more robust and accurate blind extraction of a single moving source than representative IVA-based methods for dynamic acoustic scenarios.