TDT4 Multilingual Broadcast News Speech Corpus

Item Name: TDT4 Multilingual Broadcast News Speech Corpus
Author(s): Junbo Kong, David Graff
LDC Catalog No.: LDC2005S11
ISBN: 1-58563-338-0
ISLRN: 381-876-213-869-8
Release Date: May 15, 2005
Member Year(s): 2005
DCMI Type(s): Sound
Sample Type: 2-channel
Sample Rate: 16000
Data Source(s): broadcast news
Project(s): GALE, TDT, TIDES
Language(s): English, Standard Arabic, Mandarin Chinese
Language ID(s): eng, arb, cmn
License(s): LDC User Agreement for Non-Members
Online Documentation: LDC2005S11 Documents
Licensing Instructions: Subscription & Standard Members, and Non-Members
Citation: Kong, Junbo, and David Graff. TDT4 Multilingual Broadcast News Speech Corpus LDC2005S11. Web Download. Philadelphia: Linguistic Data Consortium, 2005.
Related Works: View


TDT4 Multilingual Broadcast News Speech Corpus was developed by the Linguistic Data Consortium (LDC) with support from the DARPA TIDES (Translingual Information Detection, Extraction and Summarization) Program. This release contains the complete set of American English, Modern Standard Arabic, and Mandarin Chinese broadcast news audio used in the 2002 and 2003 Topic Detection and Tracking (TDT) technology evaluations, totalling approximately 607 hours.

The corresponding transcripts for these speech files are available in TDT4 Multilingual Text and Annotations (LDC2005T16). That release also includes newswire data and topic relevance annotations.

TDT refers to automatic techniques for finding topically related material in streams of data such as newswire and broadcast news. Evaluation tasks in 2002 and 2003 included the segmentation of a news source into stories, the tracking of known topics, the detection of unknown topics, the detection of initial stories on unknown topics, and the detection of pairs of stories on the same topic.


The TDT4 corpus contains news data collected daily from 20 news sources (13 broadcast, seven newswire) in three languages (American English, Mandarin Chinese, and Modern Standard Arabic), over a period of four months (October 2000 through January 2001). Here's a breakdown of the broadcast data included in this release with number of files and time length by source:

Language Source Program Files Hours
Arabic Voice of America Modern Standard Arabic news programs 68 67.9
Nile TV   41 20.4
Language Totals 109 88.3
English Cable News Network Headline News 129 64.5
American Broadcasting Company World News Tonight 77 38.5
National Broadcasting Company NBC Nightly News Tonight 70 35
Public Radio International The World 62 62
Voice of America English news programs 69 68.4
MS-NBC News with Brian Williams 43 43
Language Totals 450 311.3
Mandarin China Broadcasting System   103 33.9
China Television System   106 42.3
Voice of America Mandarin Chinese news programs 64 63.6
China National Radio   86 36.6
China Central Television   65 31
Language Totals 424 207.4
Totals 983 607

The files are presented in NIST SPHERE format, 1-channel, sampled at 16 kHz.


For an example of the data in this corpus, please listen to these samples:


The initial July 2005 release contained an error in which a small number of files were misnamed. That error has been corrected. Users who received the original release should have also received the correction. Anyone who ordered this publication before Nov. 1, 2005 and has not received a correction should contact LDC's Membership Office at

Available Media

View Fees

Login for the applicable fee