PMC

Excluding amplitude-related voice features improves robustness to noise. Top: MAEs without noise (circles) and with 2% (diamonds) and 5% (squares) additive noise to all voice features. Bottom: changes in MAEs due to addition of 5% noise to individual voice features.

Performance with different voice features sets. Top: MAEs without noise (circles) and with 2% (diamonds) and 5% (squares) additive noise to all voice features. Bottom: changes in MAEs due to addition of 5% noise to individual voice features.

Performance with voice features of small sensitivity excluded. Top: MAEs without noise (circles) and with 2% (diamonds) and 5% (squares) additive noise to all voice features. Bottom: changes in MAEs due to addition of 5% noise to individual voice features.

(a) The correlation integral ln C(r) versus ln (r) for the tremor voice, where the curves from bottom to top correspond to m = 1,2,…,15, respectively; (b) The correlation integral ln C(r) versus ln (r) for the surrogate data of the tremor voice, where the curves from bottom to top correspond to m = 1,2,…,15, respectively; (c) The estimated dimension versus m, in which the curves correspond to the normal voice (—□—), the low-frequency modulation tremor voice (—○—), the irregular tremor voice (—△—), the surrogate data of the irregular tremor voice (—▽—), and the random noise (—◇—), respectively.