Untersuchungen zur Konsistenz prosodischer Etikettierungen

For the automatic analysis of prosody in speech recognition systems prosodically labeled speech data are required in large amounts for test and training purposes. To achieve this, appropriate prosodic labels have to be defined that are not only capabel of describing prosody exactly but can be identified consistently even by transcribers without a profound theoretical knowledge of prosody. In a first experiment a corpus of nearly 500 utterances was labeled in parallel by untrained transcribers using a simple and basic inventory of prosodic labels. The consistency of the resulting prosodic transcriptions was investigated. (orig.)SIGLEAvailable from TIB Hannover: RR 5221(77)+a / FIZ - Fachinformationszzentrum Karlsruhe / TIB - Technische InformationsbibliothekBundesministerium fuer Bildung, Wissenschaft, Forschung und Technologie, Bonn (Germany)DEGerman