Multichannel Speech Separation with Narrow-band Conformer

This paper addresses the problem of multi-channel multi-speech separation based on deep learning techniques. In the short time Fourier transform domain, we propose an end-to-end narrow-band network that directly takes as input the multi-channel mixture signals of one frequency, and outputs the separated signals of this frequency. In narrow-band, the spatial information (or inter-channel difference) can well discriminate between speakers at different positions. This information is intensively…

Single-Channel Speech Dereverberation using Subband Network with A Reverberation Time Shortening Target

This work proposes a subband network for single-channel speech dereverberation, and also a new learning target based on reverberation time shortening. The learning target for dereverberation is usually set as the direct-path speech, which is a very challenging target and may leads to a large prediction error. This can be relaxed by leaving some early reflections in the target. No matter whether leaving early reflections, the target suddenly truncates the reverberation, which is not suitable for network training. In this work, we proposes a new learning target to suppress reverberation and meanwhile maintain the exponential decaying property of reverberation…