A Robust Acoustic Echo Canceller Based on Subband Fast Kalman Algorithm and Daul-filter Structure

Abstract

This paper proposed an acoustic echo canceller with excellent convergence performance and robust double talk detection (DTD) performance. The fast Kalman algorithm has the same order of calculation as the normalized least mean square (NLMS) algorithm but a faster convergence speed. However, it is vulnerable to instability due to finite-precision errors or quantization errors. In addition, the problem of double talk seriously affects the global performance of echo cancellation system especially for faster adaptive algorithm. In this paper, the sub-band fast Kalman algorithm is used for adaptive filtering. By utilizing the dual-filter structure, a new robust transfer method is proposed, which effectively detects the double talk and solves the problem of instability of the fast Kalman algorithm. The experimental studies using utterances corrupted by echoes prove that the proposed algorithm outperforms the Speex AEC algorithm and the algorithm using conventional transfer logic.