Workshop on ToBI for Spontaneous English Speech

点击翻译按钮获取中文摘要

With National Science Foundation support, Simmons College will host an international workshop August 13-15, 2004, on the Tones and Break Indices (ToBI) labeling system. Interacting with computers using speech is still not a natural process, and one of the reasons is that computers do not fully incorporate the prosodic information (intonation and timing) of spoken utterances. The ToBI labeling system was developed with NSF support in the 1990's to label prosodic features, such as emphasis and phrasing, which convey information vital to the meaning of speech. Since that time, the ToBI conventions have furthered the understanding of prosody by allowing a large community of researchers to share data using common labels and to extend these conventions to other languages. Most of these advances have come in theoretical areas; in contrast, applications to automatic speech synthesis, recognition and understanding systems have been slowed by the challenges of using ToBI, which was developed largely on the basis of scripted speech, to label the kind of natural spontaneous speech that is now a research priority. The workshop will address this problem by (i) identifying some of the critical labeling challenges of spontaneous speech, through preparatory on-line discussions of example utterances; (ii) organizing workshop sessions around these issues and incorporating the group's resolution of these issues (as well as a larger sampling of utterances) in a revised ToBI manual, available on-line; and (iii) providing an ongoing forum in the form of a website, hosted at Simmons College, on which challenging utterances are posted monthly for labeling by interested participants, and labeling results are summarized for on-line discussion. The extension of ToBI to meet the challenges of spontaneous speech, such as disfluencies, interruptions and unscripted dialogue structures, will expand the already-demonstrated benefits of a common prosody labeling system. Being able to share data across laboratories means faster progress toward understanding how speakers use prosody in communicative speech, because it helps investigators formulate theories of intonation and timing, design experiments for testing their theoretical predictions, and develop algorithms for dealing with prosody in automatic processing systems. As a result, this new tool will have the potential to make human-computer interactions using speech more natural. Additional benefits will include better methods for automatic translation, improved materials for second language teaching and more user-friendly aids for those with speech-related handicaps.