This example demonstrates a traditional iVector system evaluated on NIST
SRE 2016. It is based on the recipe in ../../sre10/v1/. In addition to the
standard features of the SRE10 recipe, it also demonstrates the use of data
augmentation for PLDA training.
The recipe uses the following data for system development. This is in
addition to the NIST SRE 2016 dataset used for evaluation (see ../README.txt).
Corpus LDC Catalog No.
SWBD2 Phase 1 LDC98S75
SWBD2 Phase 2 LDC99S79
SWBD2 Phase 3 LDC2002S06
SWBD Cellular 1 LDC2001S13
SWBD Cellular 2 LDC2004S07
SRE2004 LDC2006S44
SRE2005 Train LDC2011S01
SRE2005 Test LDC2011S04
SRE2006 Train LDC2011S09
SRE2006 Test 1 LDC2011S10
SRE2006 Test 2 LDC2012S01
SRE2008 Train LDC2011S05
SRE2008 Test LDC2011S08
SRE2010 Eval LDC2017S06
Mixer 6 LDC2013S03
The following datasets are used in data augmentation.
MUSAN http://www.openslr.org/17
RIR_NOISES http://www.openslr.org/28