eduzhai > Applied Sciences > Engineering >

MLNET An Adaptive Multiple Receptive-field Attention Neural Network for Voice Activity Detection

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: Voice activity detection (VAD) makes a distinction between speech andnon-speech and its performance is of crucial importance for speech basedservices. Recently, deep neural network (DNN)-based VADs have achieved betterperformance than conventional signal processing methods. The existed DNNbasedmodels always handcrafted a fixed window to make use of the contextual speechinformation to improve the performance of VAD. However, the fixed window ofcontextual speech information can t handle various unpredicatable noiseenvironments and highlight the critical speech information to VAD task. Inorder to solve this problem, this paper proposed an adaptive multiplereceptive-field attention neural network, called MLNET, to finish VAD task. TheMLNET leveraged multi-branches to extract multiple contextual speechinformation and investigated an effective attention block to weight the mostcrucial parts of the context for final classification. Experiments inreal-world scenarios demonstrated that the proposed MLNET-based modeloutperformed other baselines.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×