Improved Estimation of Parkinsonian Vowel Quality through Acoustic Feature Assimilation

Table 1

Correlation coefficients and standard deviation of prediction error (SDPE) of objective metrics with subjective scores. Indices that are bolded in a column are statistically similar in their performance but are significantly better than others in that column. Significance of correlation values indicated by and .

Metric	Full set				PCA				Reduced
	Correlation (training)	SDPE (training)	Correlation (test)	SDPE (test)	Correlation (training)	SDPE (training)	Correlation (test)	SDPE (test)	Correlation (training)	SPDE (training)	Correlation (test)	SDPE (test)
SRMR		17.26	0.16	17.74	—	—	—	—	—	—	—	—
ModA		16.48	0.33	16.97	—	—	—	—	—	—	—	—
CPP		17.20		15.24	—	—	—	—	—	—	—	—
HNR		15.01		12.09	—	—	—	—	—	—	—	—
RPDE		10.79		11.98	—	—	—	—	—	—	—	—
GFCC-LR		9.15		17.44		13.63	0.42	16.35		14.86		15.05
GFCC-SVR		14.35	0.06	17.98		14.35	0.30	17.19		15.10		15.39
LCQA-LR		10.27		16.00		11.66		15.05		11.87		11.92
LCQA-SVR		13.48		13.54		12.26		15.28		13.48		15.28
Combined-LR		8.85		15.50		12.26		12.87		10.52		10.81
Combined-SVR		11.87		11.50		12.63		13.54		12.63		10.31

PCA: principal component analysis; RPDE: recurrence period density entropy; SRMR: speech-to-reverberation modulation ratio; HNR: harmonics-to-noise ratio; CPP: cepstral peak prominence; ModA: modulation area; GFCC: Gammatone frequency cepstral coefficients; LCQA: Low Complexity Quality Assessment; and SVR: support vector regression.