eduzhai > Applied Sciences > Engineering >

Overcoming label noise in audio event detection using sequential labeling

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: This paper addresses the noisy label issue in audio event detection (AED) byrefining strong labels as sequential labels with inaccurate timestamps removed.In AED, strong labels contain the occurrence of a specific event and itstimestamps corresponding to the start and end of the event in an audio clip.The timestamps depend on subjectivity of each annotator, and their label noiseis inevitable. Contrary to the strong labels, weak labels indicate only theoccurrence of a specific event. They do not have the label noise caused by thetimestamps, but the time information is excluded. To fully exploit informationfrom available strong and weak labels, we propose an AED scheme to train withsequential labels in addition to the given strong and weak labels afterconverting the strong labels into the sequential labels. Using sequentiallabels consistently improved the performance particularly with thesegment-based F-score by focusing on occurrences of events. In themean-teacher-based approach for semi-supervised learning, including an earlystep with sequential prediction in addition to supervised learning withsequential labels mitigated label noise and inaccurate prediction of theteacher model and improved the segment-based F-score significantly whilemaintaining the event-based F-score.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×