A New Real-Time Acoustic Echo Cancellation Algorithm Using Blind Source Separation and Multi-delay Filter

摘要

Adaptive algorithm is a traditional method for solving acoustic echo cancellation (AEC) problem, which needs to perform well in different scenes, especially in the double talk (DT) scenario. It is found that some algorithms designed for blind source separation (BSS) performs well in the DT scenario, which require little prior information. However, in real-time processing, the frame length of the input should be short, otherwise the performance will be worse accordingly. In this paper, we propose an algorithm framework based on BSS and multi-delay filter (MDF) for AEC, in which the coefficients are updated by auxiliary independent vector analysis (AuxIVA) algorithm. The numerical studies using utterances corrupted by echoes under different reverberation times prove that the improved algorithm outperforms the Speex AEC algorithm.