Estimation of the parameters of the quantitative intonation model with continuous wavelet analysis

Intonation generation in state-of-the-art speech synthesis requires the analysis of a large amount of data. Therefore reliable algorithms for the extraction of the parameters of an intonation model from a given F0 contour are required. This contribution proposes improvements concerning the extraction of the parameters of the quantitative intonation model developed by Fujisaki. The improvements are mainly based on the application of the continuous wavelet transform for the detection of accents and phrases in a F0 contour. A detailed explanation of the underlying idea of this approach is given and the implemented algorithm is described. Results prove that with the proposed method a significant improvement in the accuracy of the extracted parameters is achieved. Thereby the structure and the rules of the algorithm are kept relatively simple.