eduzhai > Applied Sciences > Engineering >

Transfer Learning for Improving Singing-voice Detection in Polyphonic Instrumental Music

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: Detecting singing-voice in polyphonic instrumental music is critical to musicinformation retrieval. To train a robust vocal detector, a large dataset markedwith vocal or non-vocal label at frame-level is essential. However, frame-levellabeling is time-consuming and labor expensive, resulting there is littlewell-labeled dataset available for singing-voice detection (S-VD). Hence, wepropose a data augmentation method for S-VD by transfer learning. In thisstudy, clean speech clips with voice activity endpoints and separateinstrumental music clips are artificially added together to simulate polyphonicvocals to train a vocal non-vocal detector. Due to the different articulationand phonation between speaking and singing, the vocal detector trained with theartificial dataset does not match well with the polyphonic music which issinging vocals together with the instrumental accompaniments. To reduce thismismatch, transfer learning is used to transfer the knowledge learned from theartificial speech-plus-music training set to a small but matched polyphonicdataset, i.e., singing vocals with accompaniments. By transferring the relatedknowledge to make up for the lack of well-labeled training data in S-VD, theproposed data augmentation method by transfer learning can improve S-VDperformance with an F-score improvement from 89.5 to 93.2 .

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×