This study is a user report on modeling and predicting the F0 contour of Japanese utterances with PENTA Model and AM Theory. Through a 2640-word corpus of Japanese, the two models were compared using two versions of PENTAtrainer, as well as newly developed AMtrainer, respectively. The respective modelling and synthesis accuracies of different models are discussed with reference to the type of input they require. The satisfactory results achieved by these tools demonstrate their potential in offering a fair and direct comparison between the two models. We call for more collaborative effort in comparative modeling to achieve this goal.

Index Terms: AM Theory, PENTA Model, Pitch accent