Automatic labeling of speech synthesis corpora