1998 DARPA/NIST Continuous Speech Recognition
Broadcast News Hub-4 English Evaluation Test Material
This single CD contains the evaluation test material used in the 1998 DARPA/NIST
Continuous Speech Recognition Broadcast News Hub-4 English Benchmark Test
administered by the NIST Spoken Natural
Language Processing Group and produced by the Linguistic Data Consortium
(LDC); catalog number LDC2000S86, isbn 1-58563-172-8. In
addition, the transcripts from the evaluation and the utf.dtd used to validate
the transcripts is now included. For more complete information, see the
1998 Hub-4 Website.
Note: This CD-ROM does not contain the Human Reference and Baseline
Recognizer transcripts for the Information Extraction - Named Entity (IE-NE)
Spoke. This material was released separately prior to the start of the
IE-NE Spoke.
Note: This CD-ROM does not contain the material for the Hub-4
Non-English evaluation. It will be released separately.
Please read this document in its entirety before beginning a test.
Note that the waveform and transcript data on this disc are licensed
through the Linguistic Data Consortium
(LDC) and are subject to usage restrictions. Contact
the LDC for license agreement information.
DOCUMENTATION
Instructions
The
Hub-4 English Evaluation Specification contains
the rules and conditions for implementing the Hub-4 English tests. Note
that the new IE-NE Spoke is still somewhat in a state of flux and the dates/metrics
for IE-NE are likely to change. See the most recent version of the evaluation
specification on the 1998
Hub-4 Website for updates.
Instructions for self-scoring and preparing and submitting your results
to NIST for official scoring will be distributed in email separately.
This year the Hub-4 test includes two distinct 1.5-hour test sets. The
set1 test set contains material from 1996 news broadcasts and is
meant to be comparable in terms of epoch and selection process with the
1997 test set. The set2 test set contains material from 1998 news
broadcasts and is meant to provide more contemporary data to exercise systems'
abilities to handle new words, speakers, etc. The waveforms and related
files have been named to correspond to these two test sets.
Evaluation Schedule
The following schedule pertains to all Hub-4 tests EXCEPT the Full
IE-NE and Quasi-IE-NE Spokes. See the 1998
Hub-4 Website for details regarding the IE-NE evaluation.
September 29, 1998 |
Deadline for site commitment to participate |
October 6, 1998 |
Deadline for sites to submit Devtest results (optional) |
October 13, 1998 |
Evaluation test data to be at participating sites, test begins |
November 10, 1998 (0700 EST) |
Deadline for submission of hub primary test results |
November 18 , 1998 |
NIST releases scores for the hub primary test results |
November 20, 1998 (0700 EST) |
Deadline for submission of spoke and hub contrast test results |
November 25, 1998 |
NIST releases scores for the spoke and hub contrast test results |
TBD (February/March 1999) |
Broadcast News Workshop |
Other Documentation
The Universal Transcription Format (UTF) used to annotate/transcribe the
1998 Hub-4 reference transcripts are documented in utf1_0v2.ps
(PostScript Version) and a copy of the utf.dtd is included in the doc/
subdirectory.
TEST MATERIAL
The 1998 Hub-4 Broadcast News test material is to be used in accordance
with the Hub-4 English Evaluation Specification.
Evaluation Map Files
As in 1997, the 1998 Hub-4E Benchmark Test supports only one CSR evaluation
mode in which no partitioning information is provided. The basic
timing information required to implement the evaluation is given in the
map files, h4e_98_1.uem (set1) and h4e_98_2.uem
(set2). These files contain only the pointers to the beginning and
end of the complete test sets. No side information is provided.
Segmentation Files
Automatically generated segmentation information for each of the two test
sets is provided in the files h4e_98_1.seg
(set1) and h4e_98_2.seg (set2).
Although sites are free to use any segmentation scheme of their choice,
these files are included for the convenience of sites without access to
segmentation algorithms and were generated using the CMUseg
Version 0.5 (compressed tar archive) automatic segmentation and classification
utility. The CMUseg utility has been graciously supplied to the DARPA community
by Carnegie Mellon University for use as a common acoustic segmentation
utility.
Participants are not required to use this segmentation, or the CMUseg
utility. They have been supplied to facilitate participation in the test.
Waveform Files
This year, the test material is contained in two SPHERE-formatted waveform
files. The file h4e_98_1.sph (set1)
contains 1.5 hours of Broadcast News excerpts from 1996. The file h4e_98_2.sph
(set2) contains 1.5 hours of Broadcast News excerpts from 1998. Each
file should be separately recognized per the Hub-4
English Evaluation Specification.
Copyright
Portions Copyright 1996 by PRI-Public Radio International
Portions Copyright 1996 by ABC News
Portions Copyright 1996 Cable News Network, Inc. All Rights Reserved.
Information from the USC program 'Marketplace' contained herein is the
property of USC Radio and the University of Southern California and is
protected by copyright. Use, duplication or disclosure by you is subject
to the restrictions set forth in the user agreement and attached to the
computer readable media provided to you by the Linguistic Data Consortium
of the University of Pennsylvania. Copyright 1996 University of Southern
California. all Rights Reserved. Marketplace is produced by USC Radio at
the University of Southern Califnoria, and is distributed to public Radio
stations nationwide by PRI-Public Radio International. Marketplace is made
possible by GE, the Corporation for Public Radio, and Public Radio Stations
nationwide.
Transcript Files
The UTF-formatted reference transcriptions for the test material are included
in this publication in transcripts.
Reference STM Files
The reference STM file for the test material that were used in scoring
the test results with SCLITE is included in this publication in h4e_98.stm.
Transcript Orthography Mapping File and Software
The orthography mapping file for the test material which is used in pre-processing
the reference and system-generated transcripts using tranfilt
Version 1.9 (compressed tar archive) prior to scoring is h4e_98.glm.
For your convenience, the orthography mapping file used in last year's
evaluation is available in the file en971128.glm.
SOFTWARE
SCLITE Speech Recognition Scoring Software
The NIST SCLITE Speech Recognition Scoring
Toolkit Version 1.2 (compressed tar archive) will be used to score
the results of the Hub-4 CSR tests.
Speech Waveform Manipulation Utilities
The Hub-4 Benchmark Test waveform files are encoded using the NIST SPeech
HEader REsources (SPHERE) format and may be manipulated using the SPHERE
Version 2.6a (compressed tar archive) utilities and libraries. If you
have questions about installing or using SPHERE, you may send email to
jonathan.fiscus@nist.gov.
Note that SPHERE is currently available only for UNIX platforms.
Software Updates
Current versions of NIST software are available via the NIST
Speech Software Website
CONTACT INFORMATION
If you have questions regarding the HUB-4 data and protocols listed in
this document. NIST software, data filtering, or scoring your recognizer
output, contact jonathan.fiscus@nist.gov.
If you are interested in participating in future NIST speech recognition
tests, contact david.pallett@nist.gov.
CAVEAT
Certain commercial equipment, instruments, software, and materials are
identified on this CD-ROM in order to adequately specify experimental procedures
used. Such identification does not imply recommendation or endorsement
by the National Institute of Standards and Technology (NIST), nor does
it imply that the equipment, instruments, software, or materials identified
are necessarily the best available for the purpose.