Utterance Clustering Using Stereo Audio Channels

Table 3

Tukey HSD (FWER = 0.05).

Group 1	Group 2	Meandiff	p-adj	Lower	Upper	Reject
(a) With overlap
Mono	mstcak	−0.0018	0.8611	−0.0079	0.0043	False
Mono	Sumdif	−0.0234	0.001	−0.0295	−0.0173	True
mstack	Sumdif	−0.0216	0.001	−0.0277	−0.0155	True
hstack	Mono	0.0187	0.001	0.0126	0.0248	True
hstack	mstack	0.0169	0.001	0.0108	0.023	True
hstack	Sumdif	−0.0047	0.1973	−0.0108	0.0014	False
(b) Without overlap
Mono	mstack	−0.0237	0.001	−0.0297	−0.0178	True
Mono	Sumdif	−0.0143	0.001	−0.0203	−0.0084	True
mstack	Sumdif	0.0094	0.001	0.0034	0.0154	True
hstack	Mono	0.0205	0.001	0.0146	0.0264	True
hstack	mstack	−0.0033	0.4938	−0.0093	0.0027	False
hstack	Sumdif	0.0061	0.0425	0.0001	0.0121	True

Positive/negative meandiff means the average error rate of Group 2 is more/less than the average error rate of Group 1.