“…A model with human-selected feature extraction (HSF) using the same data split and softmax configuration was trained on several widely used manufactured features including fundamental frequency [19], pitch related features [20], energy related features [21], zero crossing rate (ZCR) [21, 22], jitter [21], shimmer [21], and Mel-frequency cepstral coefficients (MFCC) [22–24]. As suggested in [19, 20], we applied the statistical functions including Maximum, Minimum, Range, Mean, Slope, Offset, Stddev, Skewness, Kurtosis, Variance, and Median for these features.…”