eduzhai > Applied Sciences > Engineering >

SAGRNN Self-Attentive Gated RNN for Binaural Speaker Separation with Interaural Cue Preservation

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: Most existing deep learning based binaural speaker separation systems focuson producing a monaural estimate for each of the target speakers, and thus donot preserve the interaural cues, which are crucial for human listeners toperform sound localization and lateralization. In this study, we addresstalker-independent binaural speaker separation with interaural cues preservedin the estimated binaural signals. Specifically, we extend a newly-developedgated recurrent neural network for monaural separation by additionallyincorporating self-attention mechanisms and dense connectivity. We develop anend-to-end multiple-input multiple-output system, which directly maps from thebinaural waveform of the mixture to those of the speech signals. Theexperimental results show that our proposed approach achieves significantlybetter separation performance than a recent binaural separation approach. Inaddition, our approach effectively preserves the interaural cues, whichimproves the accuracy of sound localization.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×