This data set consists of 18 text files containing transcripts prepared by the LDC for Voice of America satellite radio news broadcasts in Korean. The broadcasts were recorded by the LDC at transmission time during a two week period between January 21, 2000 and February 7, 2000.
Nine of the broadcasts are 30 minutes long, and the other nine broadcasts are 60 minutes long. The file names indicate the date (YYYYMMDD)and the begin and end times (HHMM EST) of the original transmission.
The character encoding is Unicode UTF-8, and the file contents are
structured using SGML. The markup strategy used here was defined by
NIST specifically for use in transcripts of broadcast news speech.
The "docs" directory provides a working DTD file, a complete
description (in the form of a PostScript file) of the document
structure, tags and attributes, and a simple text file listing the 18 data file names in the corpus.
The transcripts have been manually time aligned at the phrasal level
and annotated to identify boundaries between news stories and speaker
turns; speaker names and gender are given where identifiable. These
annotations are all provided via the SGML tags and their attributes.
A strong effort has been made to identify all unique speakers across
the transcripts. However, there may be cases where an individual
speaker has not been recognized and has been given a unique, anonymous
Audio files for these transcripts are available as a separate corpus from the LDC: LDC2006S42.
For an example of the text in this corpus, please view this screenshot of the transcriptions.
Portions © 2000, 2006 Trustees of the University of Pennsylvania