Table 2.
Names of all the features, code source used for collecting them, and a short descriptiona.
| Feature | Code source | Short description | |||
| Pitch | |||||
|
|
MedianPitch b | Little et al [36] | Median principal frequency | ||
|
|
MeanPitch | Boersma and Weenink [33] | Mean principal frequency | ||
|
|
StdDevPitch | Little et al [36] | SD in principal frequency | ||
| Jitter | |||||
|
|
MeanJitter | Little et al [36] | Perturbation in principal frequency (mean variation) | ||
|
|
MedianJitter | Little et al [36] | Perturbation in principal frequency (median variation) | ||
|
|
LocalJitter | Boersma and Weenink [33] | Average absolute difference between consecutive periods, divided by the average period | ||
|
|
LocalAbsoluteJitter | Boersma and Weenink [33] | Average absolute difference between consecutive periods measured in seconds | ||
|
|
RapJitter | Boersma and Weenink [33] | Average absolute difference between a period and the average of it and its two neighbors | ||
|
|
Ppq5Jitter | Boersma and Weenink [33] | 5-point period perturbation quotient | ||
|
|
DdpJitter | Boersma and Weenink [33] | Difference of differences of periods of principal frequency | ||
| Shimmer | |||||
|
|
MeanShimmer | Little et al [36] | Amplitude perturbation (using mean) | ||
|
|
MedianShimmer | Little et al [36] | Amplitude perturbation (using median) | ||
|
|
LocalShimmer | Boersma and Weenink [33] | Average absolute difference between amplitudes of consecutive period | ||
|
|
LocaldbShimmer | Boersma and Weenink [33] | Average absolute base-10 logarithm of the difference between amplitudes of consecutive period | ||
|
|
Apq3Shimmer | Boersma and Weenink [33] | 3-point amplitude perturbation quotient | ||
|
|
Apq5Shimmer | Boersma and Weenink [33] | 5-point amplitude perturbation quotient | ||
|
|
Apq11Shimmer | Boersma and Weenink [33] | 11-point amplitude perturbation quotient | ||
|
|
DdaShimmer | Boersma and Weenink [33] | Shimmer calculated by difference in differences of amplitude | ||
| MFCCc | |||||
|
|
MeanMFCC (0-12) | Little et al [36] | 13 features of mean MFCC | ||
|
|
VariationMFCC (0-12) | Little et al [36] | 13 features of mean variation of MFCC | ||
|
|
RelBandPower (0-3) | Tsanas et al [39] | 4 features capturing relative band power in 4 spectrum ranges | ||
| HNR d | Boersma and Weenink [33] | Signal-to-noise ratio | |||
| RPDE e | Little et al [36] | Pitch estimation uncertainty | |||
| DFA f | Little et al [36] | Measure of stochastic self-similarity in turbulent noise | |||
| PPE g | Little et al [36] | Measure of inability of maintaining constant pitch | |||
aWe collected the features using the code or methodology described in the corresponding code-source entry.
bThe correlated features were removed, and features in italicized text were used to build the models. Feature names are preceded by the loosely defined umbrella category they belong to.
cMFCC: mel-frequency cepstral coefficient.
dHNR: harmonic-to-noise ratio.
eRPDE: recurrence period density entropy.
fDFA: detrended fluctuation analysis.
gPPE: pitch period entropy.