IfI Colloquium: Some Effects of Information Density in Speech Production and Synthesis, April 19, 2018


Prof. Dr. Bernd Möbius, Universität Saarbrücken, Germany

Date: Thursday, April 19, 2018, 17:15 h

Location: BIN 2.A.01


Language offers speakers a multitude of choices of how to encode their messages. A growing body of research indicates that measures derived from information theory correlate with aspects of human language and sppech processing. Specifically, the ease of processing a linguistic expression is correlated with the predictability of this expression in context. In this talk I will report on our recent analyses of the impact of information density on acoustic-phonetic features of spoken language in six languages: American English, Czech, Finnish, French, German, and Polish. Information density is estimated from phone-level n-gram language models trained on large corpora. Our findings suggest that the prosodic structure, i.e. phrasing and accenting, mediates between requirements of efficient communication and the speech signal. However, this mediation is not perfect, as we found evidence for additional, direct effects of changes in information density on the phonetic structure of utterances. These effects appear to be stable across languages and different speech rates. I will also summarize work in our lab on the inclusion of information-theoretic features in statistical parametric speech synthesis.
This is joint work by Bistra Andreeva, Erika Brandt, Zofia Malisz, Bernd Möbius, Yoonmi Oh, Frank Zimmerer, Ingmar Steiner, and Sébastien Le Maguer.


Prof. Dr. Bernd Möbius is professor of Phonetics and Phonology at Saarland University. He holds a Dr. degree (Ph.D.) and M.A. from the University of Bonn (1992, 1985) and was a senior research scientist at Bell Labs (1993-1998) before joining the Institute of Natural Language Processing at the University of Stuttgart (1999-2007). He was acting chair of Phonetics and Speech Communication at the University of Bonn (2007-2010). A central theme of his research is to integrate phonetic knowledge in speech technology. He has worked extensively on prosody and text-to-speech synthesis. Recent work has focused on experimental methods and computational simulations to study aspects of speech production, perception and acquisition. He was a member of the Board of International Speech Communication Association (ISCA, 2007-2015) and was a founding member and chairman of ISCA's special interest group on speech synthesis (SynSIG, 2002-2005). He is currently serving as Editor-in-Chief of the Speech Communication journal.