Abstract: This paper tackles the problem of corpus based ToBI symbol automatic identification. We focus on the binary decision of accent vs. no accent tone presence. We test a cross-lingual alternative to identify accents in a given language using supervised data learning tools trained with data of a different language. A multilayer perceptron and a C4.5 decision tree have been trained with the English Boston Radio News corpus and we test its capabilities on predicting accents in a Spanish corpus. Results are promising leading us to make a discussion of the application of previous work on ToBI accents multiclass classification.
Index Terms: prosody, ToBI, crosslingual, automatic recognition of prosody.