Table 3.
Area Under the Curve (AUC) and False Positive Rate (FPR) at a 0.90 True Positive Rate (TPR) for the seven feature sets and two classifiers: Random Forest (RF) and Multi-Layer Perceptron (MLP).
| Feature Set | AUC | FPR at 0.90 TPR | ||
|---|---|---|---|---|
| RF | MLP | RF | MLP | |
| Energy threshold | 0.715 | 0.62 | ||
| PSD | 0.813 | 0.818 | 0.53 | 0.52 |
| Baseline | 0.830 | 0.829 | 0.45 | 0.45 |
| MFCCs | 0.851 | 0.853 | 0.40 | 0.39 |
| CNN | 0.850 | 0.850 | 0.38 | 0.41 |
| MFCCs-T/NT | 0.869 | 0.870 | 0.33 | 0.33 |
| CNN-T/NT | 0.884 | 0.887 | 0.32 | 0.30 |
Note: The energy threshold constitutes a 1D feature vector, which makes RF and MLP unapplicable. AUC and FPR at a 0.90 TPR are computed by sliding the threshold over the range of possible values.