This paper proposes a method for automatic detection of major phrase boundaries as a first step toward full automatic annotation of prosodic structure. The modeling error of fundamental frequency (F0) and an F0 control modeling appropriateness measure (AIC) are employed to evaluate a prosodic major phrase boundary hypotheses. Experimental results are presented showing correct prediction of 66% of major phrase boundaries in Japanese read speech.
Bibliographic reference. Hirai, Toshio / Higuchi, Norio / Sagisaka, Yoshinori (1995): "Automatic detection of major phrase boundaries using statistical properties of superpositional F0 control model parameters", In EUROSPEECH-1995, 1341-1344.