eduzhai > Applied Sciences > Engineering >

Music SketchNet Controllable Music Generation via Factorized Representations of Pitch and Rhythm

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 8 pages

Abstract: Drawing an analogy with automatic image completion systems, we propose MusicSketchNet, a neural network framework that allows users to specify partialmusical ideas guiding automatic music generation. We focus on generating themissing measures in incomplete monophonic musical pieces, conditioned onsurrounding context, and optionally guided by user-specified pitch and rhythmsnippets. First, we introduce SketchVAE, a novel variational autoencoder thatexplicitly factorizes rhythm and pitch contour to form the basis of ourproposed model. Then we introduce two discriminative architectures,SketchInpainter and SketchConnector, that in conjunction perform the guidedmusic completion, filling in representations for the missing measuresconditioned on surrounding context and user-specified snippets. We evaluateSketchNet on a standard dataset of Irish folk music and compare with modelsfrom recent works. When used for music completion, our approach outperforms thestate-of-the-art both in terms of objective metrics and subjective listeningtests. Finally, we demonstrate that our model can successfully incorporateuser-specified snippets during the generation process.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...