NIST 2006 Speaker Recognition Evaluation - Answer Keys and Supporting Files

Answer Keys

The answer key maps the test segments to the true speakers.

• sre06_test_seg_key_v10.txt and its description


• sre06_model_key_v8.txt and its description

Submission Filter Script

NIST uses a pre-processing script (filter-submission.v4.pl) to filter out trials that were found to be in error.

• model-errors-v2.txt - The training models in this list are considered *problematic* due to their having only silence, not enough speech, etc. Any trials involving these models will be removed prior to scoring.


• testseg-errors-v5.txt - The test segments in this list are considered *problematic*. The first part contains segments that just are silence, do not have enough speech, etc. Any trials involving these test segments will be removed prior to scoring. The second part contains segments that came from the data making up the training model. The trials involving these pairs will be removed prior to scoring.