Machine Transcription of Guqin Tablature and Automatic Music Rhythm Tagging

摘要

Guqin is an ancient Chinese musical instrument of the zither family. It is an important work for guqin researchers and artist to transcribe old guqin tablature into a playable form (e.g. staves). Since guqin tablature does not indicate tempo or rhythm, the main part of the transcription is to work it out. This paper introduces the research on automatic transcription of guqin tablature, and generalizes it to the problem of music rhythm tagging. Music rhythm tagging is computing the possible rhythm information according to a series of pitch values as input parameters. We employ an n-gram model based method and compare it with the maximum entropy model and a baseline method. An average result of 65% accuracy is achieved in our experiment. This paper also discusses the difference between the evaluation criteria for the quality of music rhythm tagging and the one of part-of-speech tagging. We present DIS subjective test, which tries to determine how machine overtake human brains in a specific task given. Automatic music rhythm tagging is a good case for the research of aesthetic modeling. In the other hand, the introduction of aesthetic model could possibly bring a marked improvement for the tagging system in the preference tests.