Use of Decisio Tree to Detect Gttm Group Bou Daries

We describe σGTTM that combines the generative theory of tonal music (GTTM) and statistical learning. We previously devised exGTTM, which has accommodated the original GTTM to computer implementations. The exGTTM has adjustable parameters, these parameters have to be manually configured. Therefore, it is not perfectly suited for automation. To make complete automation possible, we combined statistical learning with GTTM to create σGTTM. To prevent the sparseness problem, we abstracted data properly by using the GTTM rules for analyzing musical structures. We use the abstracted data to construct a decision tree, which is a model of decisions and their possible consequences. With σGTTM, we can segment a melody automatically from the decision tree, dependent on conditional probability. Experimental results showed that σGTTM outperformed the baseline exGTTM.