eduzhai > Applied Sciences > Engineering >

Non-Local Musical Statistics as Guides for Audio-to-Score Piano Transcription

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 16 pages

Abstract: We present an automatic piano transcription system that converts polyphonicaudio recordings into musical scores. This has been a long-standing problem ofmusic information processing, and recent studies have made remarkable progressin the two main component techniques: multipitch detection and rhythmquantization. Given this situation, we study a method integratingdeep-neural-network-based multipitch detection and statistical-model-basedrhythm quantization. In the first part, we conducted systematic evaluations andfound that while the present method achieved high transcription accuracies atthe note level, some global characteristics of music, such as tempo scale,metre (time signature), and bar line positions, were often incorrectlyestimated. In the second part, we formulated non-local statistics of pitch andrhythmic contents that are derived from musical knowledge and studied theireffects in inferring those global characteristics. We found that thesestatistics are markedly effective for improving the transcription results andthat their optimal combination includes statistics obtained from separated handparts. The integrated method had an overall transcription error rate of 7.1 and a downbeat F-measure of 85.6 on a dataset of popular piano music, and thegenerated transcriptions can be partially used for music performance andassisting human transcribers, thus demonstrating the potential for practicalapplications.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×