15th International Congress of Phonetic Sciences (ICPhS-15)
In this paper a data-driven approach in prediction of word level prosody breaks will be presented. A semi-automatic approach of symbolic prosody raw text annotation will be introduced together with a module for prediction of annotated tags for Slovenian language. The prediction module is based on an automatic learning technique, which depends on the construction of a large corpus labeled appropriately. This labeling can be done either automatically, or by hand. While automatic labeling can be less accurate than hand labeling, the latter is very time consuming and, in many cases, inconsistent. Therefore, we constructed an interactive tool for word level prosody annotation (major/minor breaks) as well as annotation of prominent words. Our major concern was speeding up the process of manual annotation with an increase of consistency and required time effort minimization. In parallel a semi-automatic approach for determining prosody breaks and prominent words for read speech was developed with the goal to support different speakers and new languages in the future. The labeled Slovenian corpus has been used to train our phrase break prediction module, implementing a neural network (NN). Experiments for the data-driven prediction of major=minor and major/minor phrase breaks were performed. The prediction accuracy achieved marks state-of-the-art word level prosody breaks prediction for the Slovenian language and is comparable to the prediction accuracy of other approaches in which more complex NN structures or other prediction methods were applied, and a much larger corpus was used for training.
Bibliographic reference. Stergar, Janez / Horvat, Bogomir / Kačič, Zdravko (2003): "Data driven symbolic prosody modeling", In ICPhS-15, 51-54.