Table 1.
Summary of the Datasets used in Pre-training. Classifiers are pre-trained on 10.29 h audio recordings annotated with four class labels: cough, sneeze, speech and noise. The datasets do not include any COVID-19 labels.
| Type | Dataset | Sampling Rate | No of Events | Total audio | Average length | Standard deviation |
|---|---|---|---|---|---|---|
| Cough | TASK dataset | 44.1 kHz | 6000 | 91 min | 0.91 s | 0.25 s |
| Brooklyn dataset | 44.1 kHz | 746 | 6.29 min | 0.51 s | 0.21 s | |
| Wallacedene dataset | 44.1 kHz | 1358 | 17.42 min | 0.77 s | 0.31 s | |
| Google Audio Set & Freesound | 16 kHz | 3098 | 32.01 min | 0.62 s | 0.23 s | |
| Total (Cough) | — | 11 202 | 2.45 h | 0.79 s | 0.23 s | |
| Sneeze | Google Audio Set & Freesound | 16 kHz | 1013 | 13.34 min | 0.79 s | 0.21 s |
| Google Audio Set & Freesound + SMOTE | 16 kHz | 9750 | 2.14 h | 0.79 s | 0.23 s | |
| Total (Sneeze) | — | 10 763 | 2.14 h | 0.79 s | 0.23 s | |
| Speech | Google Audio Set & Freesound | 16 kHz | 2326 | 22.48 min | 0.58 s | 0.14 s |
| LibriSpeech | 16 kHz | 56 | 2.54 h | 2.72 min | 0.91 min | |
| Total (Speech) | — | 2382 | 2.91 h | 4.39 s | 0.42 s | |
| Noise | TASK dataset | 44.1 kHz | 12 714 | 2.79 h | 0.79 s | 0.23 s |
| Google Audio Set & Freesound | 16 kHz | 1027 | 11.13 min | 0.65 s | 0.26 s | |
| Total (Noise) | — | 13 741 | 2.79 h | 0.79 s | 0.23 s |