2001 NIST Speaker Recognition Evaluation Corpus
|Item Name:||2001 NIST Speaker Recognition Evaluation Corpus|
|Author(s):||Mark Przybocki, Alvin Martin|
|LDC Catalog No.:||LDC2002S34|
|Release Date:||September 26, 2002|
|Data Source(s):||telephone speech|
|Application(s):||speaker identification, speaker segmentation and tracking|
LDC User Agreement for Non-Members
|Online Documentation:||LDC2002S34 Documents|
|Licensing Instructions:||Subscription & Standard Members, and Non-Members|
|Citation:||Przybocki, Mark, and Alvin Martin. 2001 NIST Speaker Recognition Evaluation Corpus LDC2002S34. Web Download. Philadelphia: Linguistic Data Consortium, 2002.|
2001 NIST Speaker Recognition Evaluation Corpus was developed by the Linguistic Data Consortium (LDC) and the National Institute of Standards and Technology (NIST). It contains 26 hours of conversational cellular telephone speech in English collected by LDC used for development and evaluation data in the NIST-sponsored 2001 Speaker Recognition Evaluation.
This release is part of an ongoing series of yearly evaluations conducted by NIST. These evaluations provide an important contribution to the direction of research efforts and the calibration of technical capabilities. They are intended to be of interest to all researchers working on the general problem of text independent speaker recognition. To this end the evaluation was designed to be simple, to focus on core technology issues, to be fully supported, and to be accessible.
The 2001 evaluation included four tasks:
- One-speaker detection
- Two-speaker detection
- Speaker Tracking
- Speaker segmentation
It also included four distinct combinations of parameters and data, the first being a repeat of 2000's evaluation using all four tasks, and the other three (which includes a test using the Spanish-language corpus AHUMADA from 2000's evaluation) exercising only the one-speaker detection task.
The files are divided into evaluation and development data. There are a total of 2,350 compressed speech files all of which are in sphere format. The sphere files are compressed and encoded in one channel 8-bit mulaw, for a total of 575,337,198 bytes (548.7 MB).
The evaluation data is divided into evaluation training data and evaluation test data. The training data consists of 174 speech files that are two minutes long. The test data comprises 2,038 speech files of varying lengths not exceeding 60 seconds.
The development data is similarly divided into development training data and development test data. The training data comprises 60 speech files with durations of two minutes per target speaker. The 78 development test data files contain segments of varying length not exceeding 60 seconds.
For an example of the data in this corpus, please listen to this sample (SPH).
None at this time.