. Author manuscript; available in PMC: 2015 Mar 1.

Published in final edited form as: Comput Speech Lang. 2012 Oct 4;28(2):10.1016/j.csl.2012.09.004. doi: 10.1016/j.csl.2012.09.004

Table 5.

Unweighted average recall for iterative and background speaker normalization when compared to no speaker normalization (Raw) and global speaker normalization on the official and matched development sets.

	Official Development Set			Matched Development Set
	Raw	Global	Iterative	Raw	Global	Iterative	Background
All	0.641	0.719	0.684	0.588	0.697	0.676	0.685
Address	0.737	0.761	0.760	0.665	0.682	0.695	0.690
Command & Control	0.660	0.702	0.677	0.564	0.679	0.663	0.671
Dialogue	0.604	0.765	0.768	0.590	0.752	0.729	0.713
Monologue	0.638	0.768	0.759	0.632	0.751	0.644	0.512
Digit String	0.654	0.747	0.729	0.583	0.701	0.648	0.654
Tongue Twister	0.711	0.783	0.752	0.591	0.727	0.681	0.682
All-divided	0.671	0.744	0.724	0.592	0.706	0.667	0.672