eduzhai > Applied Sciences > Engineering >

Score-informed Networks for Music Performance Assessment

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 8 pages

Abstract: The assessment of music performances in most cases takes into account theunderlying musical score being performed. While there have been severalautomatic approaches for objective music performance assessment (MPA) based onextracted features from both the performance audio and the score, deep neuralnetwork-based methods incorporating score information into MPA models have notyet been investigated. In this paper, we introduce three different modelscapable of score-informed performance assessment. These are (i) a convolutionalneural network that utilizes a simple time-series input comprising of alignedpitch contours and score, (ii) a joint embedding model which learns a jointlatent space for pitch contours and scores, and (iii) a distance matrix-basedconvolutional neural network which utilizes patterns in the distance matrixbetween pitch contours and musical score to predict assessment ratings. Ourresults provide insights into the suitability of different architectures andinput representations and demonstrate the benefits of score-informed models ascompared to score-independent models.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×