TABLE V.

Cross-utterance classification accuracy on mixtures involving a novel utterance. Data were from exp. 3. Same-talker models were trained on two utterances, different-talker models on three. Results are shown with (+) and without (−) aligning the clean utterances to a reference. All results are significantly better than chance at a 0.05 level according to a one-sided binomial test.

Talker	Warp	/ɑtʃɑ/	/ɑdʒɑ/	/ɑdɑ/	/ɑtɑ/	/ɑfɑ/	/ɑvɑ/	Avg.
Same	+	65.2	62.7	74.1	74.4	67.0	63.5	67.8
Same	−	63.6	63.9	63.4	74.0	57.2	64.5	64.4
Diff.	+	62.3	58.7	69.3	73.9	59.9	61.6	64.3
Diff.	−	55.7	58.9	66.8	63.1	61.6	59.9	61.0