Decision tree based duration prediction in Mandarin

This paper reports the methodology and results of decision tree based duration prediction for Mandarin text-to-speech system developed by the FUJITSU Laboratories. Syllable initials and finals are the basic units in our duration study. In this paper, the factors that influence the finals such as phrase boundary and phone context are discussed in detail. Experiments show that the prosodic factor of whether the right phrase boundary level is prosodic word or higher level is the most important determinant of duration. Furthermore, the degree of phrase boundary vowel lengthening may vary depending on the different kinds of finals. And this paper also explains the methods for objective evaluation of the performance of the duration prediction model.