Enhancing the Separation of Sources in Close Proximity and Intense Reverberant Conditions by Integrating Spatial Cues with Spatial Covariance

摘要

While many source separation algorithms based on spatial cues perform well in low and medium reverberant conditions when the separation between the sources is large, their performance decline sharply in highly reverberant situations and even more, when the sources are placed very close to each other. In this paper, we have proposed a binaural time frequency masking source separation algorithm which integrates the spatial cues extracted from the audio mixtures with the spatial covariance that exists among the two mixture channels to achieve improved signal to distortion ratio (SDR) for sources placed in close proximity in highly reverberant environment.