Speech Prosody 2004
This paper discusses the problems in the automatic method to determine the discrete parameters of the proposed F0 model from the speech wave. The dynamic programming (henceforth DP method) and the least mean square error (LMSE) methods serve in the two-step algorithm proposed in this paper. Furthermore, in order to automatically detect the optimal number of phrase commands, decrease of LMSE is used. From the experiment results on a set of 11 sentences spoken by four Japanese speakers, we obtained 84.1% correct rate of phrase component extraction.
Bibliographic reference. Bu, Shehui / Yamamoto, Mikio / Itahashi, Shuichi (2004): "Evaluation of a method for automatic determination of f0 model parameters", In SP-2004, 431-434.