The Rich Transcription 2004 Spring Meeting Recognition Evaluation

Published

April 1, 2004

Author(s)

John S. Garofolo, Christophe Laprun, Jonathan G. Fiscus

Abstract

This paper presents the design and results of the Rich Transcription 2004 Spring Meeting Recognition Evaluation. The evaluation included both Speaker Segmentation (SPKR) and Speech-to-Text Transcription (STT) tasks. Three microphone conditions were supported: - Multiple Distant Microphones (the primary condition of interest), - Single Distant Microphone (SDM), and - Individual Head Microphones (IHM) (for the STT task only). The 3 microphone conditions permitted the examination of performance for distant vs. close-talking microphones and single vs. multiple distant microphones. Multi-site training and development corpora were provided to the evaluation participants. The evaluation test set consisted of 8 11-minute meeting excerpts collected at CMU, ICSI, the LDC, and NIST and transcribed by the LDC. Because meetings contain a great deal of overlapping/spontaneous speech, the evaluation featured a new experimental scoring of overlapping speech for the STT task.

Citation

OTHER -

NIST Pub Series

Other

Pub Type

NIST Pubs

Keywords

evaluation, meeting domain, rich transcription, speech recognition, spontaneous speech

Citation

Garofolo, J. , Laprun, C. and Fiscus, J. (2004), The Rich Transcription 2004 Spring Meeting Recognition Evaluation, Other, National Institute of Standards and Technology, Gaithersburg, MD (Accessed December 11, 2024)