Automatic Evaluation of Voice Quality Using Text-Based Laryngograph Measurements and Prosodic Analysis

Table 5

Best feature sets for human-machine correlation and their weights in the regression formulae.

Feature	Context		w/o CFx		w/o CFx
DurNorm	WPW					0.377		0.499	0.378
DurNorm	W						0.513			0.402
F0Min	W
F0Mean	W
F0Onset	W					0.173
F0OffPos	W						0.322	0.120	0.185	0.236
EnNorm	WPW									0.343
EnNorm	W								0.155
MeanJitter	15 W	0.118	0.186	0.113	0.249	0.239	0.366	0.368	0.320	0.208
MeanShimmer	15 W	0.144	0.138	0.145	0.114
StandDevShimmer	15 W
#+Voiced	15 W	0.321	0.347	0.334	0.324		0.094			0.122
RelNum+/−Voiced	15 W							0.218	0.082
CFx		0.210		0.206
CQx							0.643	0.495		0.506
		0.71	0.66	0.71	0.67	0.36	0.53	0.47	0.45	0.49
ρ		0.57	0.49	0.58	0.49	0.27	0.54	0.46	0.45	0.55
Significance level		<0.001	<0.001	<0.001	<0.001	0.003	<0.001	<0.001	<0.001	<0.001

Contexts: W: word, WPW: word-pause-word, 15 W: 15 words (“global” feature). The correlations of the respective set to the human reference are given by (Pearson) and (Spearman).