We propose a method for syllabic stress annotation which does not require manual labels for the learning process, but uses stress labels automatically generated from a multiscale model of rhythm perception. The model outputs a sequence of events, corresponding to the sequences of strong-weak syllables present in speech, based on which a stressed/unstressed decision is taken. We tested our approach on two languages, Catalan and Spanish, and we found that a classifier employing the automatic labels for learning improves performance over the base-line for both languages. We also compared the results of this system with those of an identical learning algorithm, but which employs manual labels for stress, as well as to the results of a clustering algorithm using the same features. It showed that the system employing automatic labels has a performance close to the one using manual labels, with both classifiers outperforming the clustering algorithm.
展开▼
