Melody expectation method based on GTTM and TPS

A method that predicts the next notes is described for assisting musical novices to play improvisations. Melody prediction is one of the most difficult problems in musical information retrieval because composers and players may or may not create melodies that conform to our expectation. The development of a melody expectation method is thus important for building a system that supports musical novices because melody expectation is one of the most basic skills for a musician. Unlike most previous prediction methods, which use statistical learning, our method evaluates the appropriateness of each candidate note from the view point of musical theory. In particular, it uses the concept of melody stability based on the generative theory of tonal music (GTTM) and the tonal pitch space (TPS) to evaluate the appropriateness of the melody. It can thus predict the candidate next notes not only from the surface structure of the melody but also from the deeper structure of the melody acquired by GTTM and TPS analysis. Experimental results showed that the method can evaluate the appropriateness of the melody sufficiently well.


  • Masatoshi Hamanaka, Keiji Hirata, Satoshi Tojo: Melody Expectation Method Based on GTTM and TPS, Proceedings of the 9th International Conference on Music Information Retrievalconference(ISMIR2008),pp.107-112, September 2008. [PDF]
  • 浜中雅俊, 平田圭二, 東条敏: “FATTAに基づくメロディ予測システム”, 情報処理学会 音楽情報科学研究会 研究報告 2008-MUS-76-9, Vol. 2008, No. 78, pp. 45-50, August 2008. [PDF]