eduzhai > Applied Sciences > Engineering >

audioLIME Listenable Explanations Using Source Separation

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: Deep neural networks (DNNs) are successfully applied in a wide variety ofmusic information retrieval (MIR) tasks but their predictions are usually notinterpretable. We propose audioLIME, a method based on Local InterpretableModel-agnostic Explanations (LIME) extended by a musical definition oflocality. The perturbations used in LIME are created by switching on offcomponents extracted by source separation which makes our explanationslistenable. We validate audioLIME on two different music tagging systems andshow that it produces sensible explanations in situations where a competingmethod cannot.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×