For training and testing data, we used the Boston University FM Radio corpus, speaker f2b [4]. This database consists of 126 utterances of single speaker female American English news-reader speech (about 45 minutes). The utterances were divided into training and test sets, with the test set comprising one quarter of the utterances. The database is labelled with segment, syllable and word boundaries including lexical stress markings. The database is also already hand-labelled with ToBI intonation labels. For our work the database was additionally labelled with Tilt intonation labels. As our automatic Tilt event labeller is still under development, the Tilt events were derived from the existing ToBI labels. In addition, the database was also fully hand labelled with Tilt events.