Table 1.
Quantitative evaluation of the different strategies on speech emotion recognition. In bold, the best model.
| TL Strategy | Inputs | Models | With VAD (InaSpeech) |
Accuracy ± 95% CI |
|---|---|---|---|---|
| - | - | Human perception [18] |
- | 67.00 |
| - | - | ZeroR | - | 13.33 ± 2.06 |
| Feature Extraction |
Deep-Spectrum embs. from fc7 of AlexNet |
SVC | No | 43.32 ± 2.56 |
| Yes | 45.80 ± 2.57 | |||
| PANNs embs. from CNN-14 |
SVC | No | 39.73 ± 2.53 | |
| Yes | 37.22 ± 2.50 | |||
| Fine Tuning |
Mel spectrograms | AlexNet | No | 60.72 ± 2.52 |
| Yes | 61.67 ± 2.51 | |||
| Mel spectrograms | CNN-14 | No | 76.58 ± 2.18 | |
| Yes | 75.25 ± 2.23 |