Table 5.
The Mean Testing Detection Results of GMM+HMM with MFCCs, k-NN with Hu Moments, 5-layer CNN with Mel-scaled Spectrograms, BiLSTM, C-BiLSTM and C-BiLSTM with Boundary Regression on Corp Dataset over 3 Runs.
| Class. | SENS(%) | SPEC(%) | ACC(%) | MCC(%) | PPV(%) | NPV(%) | IoU |
|---|---|---|---|---|---|---|---|
| Prior Works | |||||||
| GMM+HMM with MFCCs [9] | 66.12 | 99.66 | 99.28 | 67.48 | 69.63 | 99.6 | 0.66 |
| k-NN with Hu Moments [13] | 63.89 | 99.72 | 99.31 | 67.87 | 72.92 | 99.58 | 0.72 |
| 5-layer CNN with Mel-scaled Spectrograms [17] | 78.54 | 99.78 | 99.53 | 79.25 | 80.47 | 99.75 | 0.79 |
| Our Works | |||||||
| Deep BiLSTM (Method 1) | 74.83 | 99.76 | 99.48 | 76.51 | 78.84 | 99.71 | 0.73 |
| C-BiLSTM (Method 2) | 82.76 | 99.77 | 99.57 | 81.53 | 80.74 | 99.8 | 0.75 |
| C-BiLSTM with boundary regression (Method 3) | 84.13 | 99.82 | 99.64 | 83.91 | 84.06 | 99.81 | 0.89 |