Informed FastICA: Semi-Blind Minimum Variance Distortionless Beamformer

TL;DR

This paper introduces a semi-blind FastICA variant that integrates MVDR beamforming, combining model-based and learning-based methods for improved blind source extraction with fast convergence.

Contribution

It replaces the orthogonality constraint in FastICA with an MVDR-based constraint, creating a semi-blind extraction method that leverages side information from covariance matrices.

Findings

Fast convergence observed in simulations

Promising performance in speaker extraction tasks

Connects blind and learning-based extraction methods

Abstract

Non-Gaussianity-based Independent Vector Extraction leads to the famous one-unit FastICA/FastIVA algorithm when the likelihood function is optimized using an approximate Newton-Raphson algorithm under the orthogonality constraint. In this paper, we replace the constraint with the analytic form of the minimum variance distortionless beamformer (MVDR), by which a semi-blind variant of FastICA/FastIVA is obtained. The side information here is provided by a weighted covariance matrix replacing the noise covariance matrix, the estimation of which is a frequent goal of neural beamformers. The algorithm thus provides an intuitive connection between model-based blind extraction and learning-based extraction. The algorithm is tested in simulations and speaker ID-guided speaker extraction, showing fast convergence and promising performance.