Mari Ostendorf has advanced the field of speech and language processing by combining pioneering advances in statistical modelling and machine learning with an understanding of the deeper structures in speech. In acoustic modeling, she developed stochastic segment models to overcome the limitations of hidden Markov models, allowing the use of richer feature sets. Ostendorf helped standardize the method for annotating prosodic events (the properties of syllables and larger units of speech such as intonation, tone, and rhythm), as a major contributor to the now standard “ToBI” method for labeling speech prosody, and demonstrated its utility for improving spoken language technology. She also created novel frameworks of language modeling that are effective in separating different contextual factors, significantly impacting language modeling for speech recognition.
An IEEE Fellow, Ostendorf is a professor of electrical engineering at the University of Washington, Seattle, WA, USA.