Parameter extraction of a quantitative intonation model with wavelet analysis and evolutionary optimization

State-of-the-art speech technology requires computation of large amounts of data. This is only possible with reliable algorithms for automatic data analysis. The paper deals with the automatic extraction of the parameters of a quantitative intonation model developed by H. Fujisaki and his coworkers (Joint Conf. of SNLP and Oriental COCOSDA, 2002). For detection of the accent and phrase commands of this model, a frequency analysis based on wavelet transforms is proposed. Furthermore, an evolutionary strategy is used to optimize the model parameters of the obtained first-order approximation. First results show that the quality of the extracted parameters is comparable to reference data.