2003 NIST Language Recognition Evaluation

Item Name: 2003 NIST Language Recognition Evaluation
Author(s): Alvin Martin, Mark Pryzbocki
LDC Catalog No.: LDC2006S31
ISBN: 1-58563-364-X
ISLRN: 610-601-655-546-3
Release Date: June 15, 2006
Member Year(s): 2006
DCMI Type(s): Sound
Sample Type: ulaw
Sample Rate: 8000
Data Source(s): telephone conversations
Project(s): NIST LRE
Application(s): speech recognition
Language(s): Vietnamese, Tamil, Spanish, Iranian Persian, Korean, Japanese, Hindi, French, English, German, Mandarin Chinese, Egyptian Arabic
Language ID(s): vie, tam, spa, pes, kor, jpn, hin, fra, eng, deu, cmn, arz
License(s): LDC User Agreement for Non-Members
Online Documentation: LDC2006S31 Documents
Licensing Instructions: Subscription & Standard Members, and Non-Members
Citation: Martin, Alvin, and Mark Pryzbocki. 2003 NIST Language Recognition Evaluation LDC2006S31. Web Download. Philadelphia: Linguistic Data Consortium, 2006.

Introduction

The goal of the NIST Language Recognition Evaluation (LRE) is to establish the baseline of current performance capability for language recognition of conversational telephone speech and to lay the groundwork for further research efforts in the field. The series had its first evaluation in 1996. 2003 NIST Language Recognition Evaluation (LRE-03) was part of this ongoing series of evaluations of language recognition technology.

Further information regarding this evaluation may be found on the 2003 NIST Language Recognition Evaluation website and in the NIST 2003 evaluation plan.

The task evaluated was the detection of a given target language. Given a test segment of speech, a target language was assigned as a test hypothesis, and the task was to determine whether this test hypothesis was true or false. This release contains both the 1996 and 2003 NIST Language Recognition Evaluations.

Data

Each speech file is one side of a "four wire" telephone conversation represented as 8-bit, 8kHz mulaw data. There are 11,830 speech files in sphere(.sph) format for a total of around forty six hours of speech. The speech data was compiled from the LDC's CALLFRIEND, CALLHOME, and Switchboard-2 corpora. Each file contains one test segment. The test segments are divided into three-second, ten-second, and thirty-second tests, each in its own directory.

Samples

For an example of the data in this corpus, please listen to this audio sample.

Updates

A typo was fixed in the index.html file. There are 11,830 sphere files, not 11,839. The updated index file is available in the online docs folder.

Available Media

View Fees

Member
Non-Member
Reduced-License
Extra Copy
Login for the applicable fee