NIST 2017 Language Recognition Evaluation

The 2017 NIST language recognition evaluation (LRE17) is the 8th cycle in an on-going language recognition evaluation series that began in 1996. The objectives of the evaluation series are (1) to explore promising new ideas in language recognition, (2) to support the development of advanced technology incorporating these ideas, and (3) to measure the performance of the current state of technology. Targeting researchers working on the general problem of text-independent, speaker-independent language recognition, the evaluation is designed to focus on core technology issues and to be simple and accessible to those wishing to participate.

LRE17 will be organized in a similar manner as LRE15, with a few differences. A main difference is in the test data used in the evaluation. In addition to conversational telephone speech (CTS), speech excerpts extracted from video data (i.e., VAST corpus) will be used as test material. Samples of the VAST data will be distributed to participants for system training and development purposes. In addition, an alternative performance measure based on cross-entropy is introduced in LRE17. The alternative measure, which is called Normalized Cross-Entropy (aka confidence score), has been used by NIST since late 90's to evaluate performance of speech-to-text systems.

Normalized Cross-Entropy

There is no cost to participate in LRE17, but participating teams must be represented at the evaluation workshop to be held in Orlando, FL, December 12-13, 2017. More information about the evaluation and registration can be found in LRE17 plan.

LRE17 Evaluation Plan(updated September 29, 2017)

Send comments to lre_poc [at] nist.gov (lre_poc[at]nist[dot]gov). A notice will go out to the mailing list and the web page will be updated when the evaluation plan becomes available.
Send email address to lre_poc [at] nist.gov (lre_poc[at]nist[dot]gov) with request to be added to the mailing list to receive automatic updates.