DISTBIC: a speaker-based segmentation for audio data indexing
Speech Communication - Special issue on accessing information in spoken audio
Prosody-based automatic segmentation of speech into sentences and topics
Speech Communication - Special issue on accessing information in spoken audio
Story segmentation of brodcast news in English, Mandarin and Arabic
NAACL-Short '06 Proceedings of the Human Language Technology Conference of the NAACL, Companion Volume: Short Papers
Hi-index | 0.00 |
In this paper we present results from a study seeking to distinguish "unprepared" from "prepared" speech in broadcast news media. The idea is to explore the results from a previous experiment concerning the characterization of filled pauses and extensions, extending the analysis of such hesitation phenomena to large audio corpus. Daily news broadcasts of Portuguese television were segmented and labeled manually in terms of several speech styles, over a range of background environments. An automatic detection of filled pauses and extensions in this audio data allowed us to correlate the presence of hesitation events with segments of unprepared speech. Distinguishing unprepared speech from prepared speech is of considerable practical interest for audio segmentation, speech processing and linguistic research. The long-term objective of this work is to automatically segment all audio genres and speaking styles as well as identify prosodic and linguistic features of the speech segments.