No abstract

Major syntactic boundaries are often accompanied by a rise in the phrase component of the fundamental frequency (F0) contour. Detecting such rises, therefore, can be signicantly helpful to the speech recognition process. We developed a method to detect syntactic boundaries with phrasecomponent rise (henceforth, phrase boundaries), based on the compression of the accent component of the F0 contour (in logarithmic scale), using a low-pass lter. In this method, F0 contours are viewed as signals in the time domain, which can be roughly separated into phrase and accent components due to their dierent frequency contents. Phrase boundaries are detected whenever a signicant rise occurs in the derivative of the ltered F0 contour. (The concepts of phrase and accent components can be found in [1]).The method managed to detect about 77% of manually detectable phrase boundaries, though with a relatively high insertion rate. The insertion rate can be reduced by using the partial AbS method, proposed by the authors [7].

No abstract

No abstract

A mode-constrained corpus-based synthesis strategy was developed for fundamental frequency (F 0 ) contours of Japanese sentences. In the training phase, the relationship between linguistic factors and the command values (amplitudes and locations) of F 0 contour generation process model was learned for a prediction module; a neural network in the current paper. Input parameters consist of linguistic information related to accentual phrases that can be automatically driven from text, such as the position of the accentual phrase in the utterance, number of morae, accent type, and morphological information. In the synthesis phase, the prediction module is used to generate command values of the model. The synthesis method was also realized based on multiple linear regression analysis to examine how each input parameter contributes to the F 0 contour generation. The use of the parametric model restricts the degrees of freedom of the mapping between linguistic and prosodic features, and thus enables to generate appropriate values even with limited training data. Experimental results showed that the method could generate F 0 contours quite close to those by the rulebased method.