Sixth European Conference on Speech Communication and Technology
We have been developing a reliable method for prosodic word boundary detection for Japanese continuous speech based on the discrete hidden Markov modeling of fundamental frequency (F0 ) contours in mora unit. Although a favorable result was obtained for ATR continuous speech corpus as reported already, experiments were done only on closed conditions. This paper reports the results on open and speaker-independent cases using database by two speakers. On average, detection rate reached around 76.0% with insertion error rate of 18.4%. Degradation from the closed condition experiment was only a little, showing the validity of the method for open conditions.
Full Paper (PDF) Gnu-Zipped Postscript
Bibliographic reference. Iwano, Koji (1999): "Prosodic word boundary detection using mora transition modeling of fundamental frequency contours -speaker independent experiments-", In EUROSPEECH'99, 231-234.