Table 7.

Unweighted average recall for all five subsystems on the development (D) and test (T) datasets, official and matched, and the accuracies obtained from fusion of the subsystems. Results submitted for the Challenge are listed as “Prev., Test Set Optimized”. All results on the GMM subsystems are for global speaker normalization.

Train Data	Test Data	Global-1	Iterative-2	[1,2]	MGMM-3	UWPP-4	[3,4]	LFA-5	[3-5]	[1-5]
Official	D,Official	0.719	0.684	0.724	0.677	0.659	0.704	0.700	0.711	0.751
Matched	D,Matched	0.697	0.676	0.699	0.626	0.639	0.658	0.656	0.668	0.717
Official	T,Official	0.699	0.690	0.697	0.632	0.615	0.656	0.619	0.651	0.688
Matched	T,Official	0.699	0.687	0.704	0.641	0.629	0.654	0.637	0.662	0.714
Prev., Test Set Optimized				0.681						0.705