$$\sigma $$GTTM III: Learning-Based Time-Span Tree Generator Based on PCFG
Masatoshi Hamanaka, Keiji Hirata, Satoshi Tojo · Lecture notes in computer science · 2016
An automatic analyzer based on the generative theory of tonal music (GTTM) for acquiring a time-span tree is described. Although an analyzer based on GTTM was previously reported, it requires manually manipulating 46 adjustable parameters on a computer screen in order to analyze a time-span tree properly. We reformalized the time-span reduction in GTTM on the basis of a probabilistic model called probabilistic context-free grammar, which enables acquiring the most likely time-span tree. Applying leave-one-out cross validation over 300 datasets revealed that the new analyzer outperformed our previously developed GTTM analyzer.