Second International Conference on Spoken Language Processing (ICSLP'92)

Banff, Alberta, Canada
October 13-16, 1992

TOBI: A Standard for Labeling English Prosody

Kim Silverman (1), Mary Beckman (2), John Pitrelli (1), Mori Ostendorf (3), Colin Wightman (4), Patti Price (5), Janet Pierrehumbert (6), Julia Hirschberg (7)

(1) NYNEX Science & Technology, Inc.; (2) Ohio State University; (3) Boston University; (4) New Mexico Institute of Mining and Technology; (5) SRI International; (6) Northwestern University; (7) AT&T Bell Laboratories, USA

An understanding of prosody is critical in basic research in speech and natural language processing, and in the technology for building high quality speech synthesis and spoken language understanding systems. Sufficient understanding and development of computational models require large amounts of prosodically transcribed speech. Unfortunately there is no single standard for prosodic transcription that is analogous to IPA for phonetic segments. To meet this need, a group of researchers with expertise in a variety of approaches to prosodic analysis and speech technology have developed TOBI: an agreed transcription system which builds on much recent progress in prosodic modelling. In a study with twenty transcribers with varied experience using this system and a total of 20,000 decisions, high inter-transcriber reliability was achieved. We report this and other evaluations of TOBI which document the consistency with which it can be used. We propose this system as a standard for prosodic transcription of large speech corpora.

Full Paper

Bibliographic reference.  Silverman, Kim / Beckman, Mary / Pitrelli, John / Ostendorf, Mori / Wightman, Colin / Price, Patti / Pierrehumbert, Janet / Hirschberg, Julia (1992): "TOBI: a standard for labeling English prosody", In ICSLP-1992, 867-870.