Heavy-tailed Distributions-Based Online Semi-blind Source Separation for Nonlinear Echo Cancellation

Recently proposed semi-blind source separation (SBSS) based acoustic echo cancellation (AEC) algorithms have attracted significant research interest due to their ability to track dynamic acoustic environments in the presence of near-end signals. Two source models are considered in these existing algorithms, i.e., the spherical generalized super-Gaussian distribution and the circular super-Gaussian distribution with a low-rank spectrogram model. In this paper, we aim to further enhance AEC performance by leveraging more flexible source models. Several novel algorithms are subsequently proposed. Simulations demonstrate the superiority of proposed algorithms in various situations.