Nonlinear residual echo suppression based on dual-stream DPRNN

Abstract The acoustic echo cannot be entirely removed by linear adaptive filters due to the nonlinear relationship between and far-end signal. Usually, a post-processing module is required further suppress echo. In this paper, we propose residual suppression method based on modification of dual-path recurrent neural network (DPRNN) improve quality speech communication. Both signal auxiliary signal, or output filter, obtained from cancelation are adopted form dual-stream for DPRNN. We validate efficacy proposed in notoriously difficult double-talk situations discuss impact different signals performance. also compare performance time domain time-frequency processing. Furthermore, an efficient applicable way deploy our off-the-shelf loudspeakers fine-tuning pre-trained model with little recorded-echo data.