On End-to-end Multi-channel Time Domain Speech Separation in Reverberant Environments
This addresses speech separation for applications in noisy, reverberant settings, offering incremental improvements through a novel network structure and dereverberation pre-processing.
The paper tackles multi-channel speech separation in reverberant environments by proposing a fully-convolutional neural network that directly separates speech from microphone recordings without conventional spatial features, achieving over 13% improvement in source separation and over 50% reduction in word error rate compared to a reference system.
This paper introduces a new method for multi-channel time domain speech separation in reverberant environments. A fully-convolutional neural network structure has been used to directly separate speech from multiple microphone recordings, with no need of conventional spatial feature extraction. To reduce the influence of reverberation on spatial feature extraction, a dereverberation pre-processing method has been applied to further improve the separation performance. A spatialized version of wsj0-2mix dataset has been simulated to evaluate the proposed system. Both source separation and speech recognition performance of the separated signals have been evaluated objectively. Experiments show that the proposed fully-convolutional network improves the source separation metric and the word error rate (WER) by more than 13% and 50% relative, respectively, over a reference system with conventional features. Applying dereverberation as pre-processing to the proposed system can further reduce the WER by 29% relative using an acoustic model trained on clean and reverberated data.