Complex angular central Gaussian mixture model for directional statistics in mask-based microphone array signal processing

Microphone array signal processing based on time-frequency masks has been applied successfully to various tasks including source separation, denoising, source localization, and source counting. Aiming to improve the performance of these techniques, here we propose a mask estimation method based on a complex Angular Central Gaussian Mixture Model (cACGMM) for multichannel observed signals. Compared to a conventional complex Watson Mixture Model (cWMM), the proposed cACGMM can model not only rotationally symmetrical but also elliptical distributions. Therefore, the cACGMM can better approximate the distribution of observed data, which is generally not rotationally symmetrical. In source separation simulations with real recorded impulse responses, the cACGMM resulted in an average 1.2 dB improvement of the Signal-to-Distortion Ratio (SDR) over the cWMM.