Spoken Language Understanding on Fluent Speech Commands (test)
99.26Accuracy3D-CNN+LSTM+CE
Evaluation Results
| Method | Links | |
|---|---|---|
| 3D-CNN+LSTM+CEloss_function=CE2021.06 | 99.26 | |
| CNN+GRU(SotA)2021.06 | 99.1 | |
| No unfreezingTraining data size=Full dataset, Pre-training=ASR-based, Unfreezing strategy=None2019.04 | 98.8 | |
| RNN+Pre-training2021.06 | 98.8 | |
| Unfreeze word layersTraining data size=Full dataset, Pre-training=ASR-based, Unfreezing strategy=Word layers2019.04 | 98.7 | |
| No unfreezingTraining data size=10% subset, Pre-training=ASR-based, Unfreezing strategy=None2019.04 | 97.9 | |
| Unfreeze word layersTraining data size=10% subset, Pre-training=ASR-based, Unfreezing strategy=Word layers2019.04 | 97.9 | |
| CNN+Segment pooling2021.06 | 97.8 | |
| Unfreeze all layersTraining data size=Full dataset, Pre-training=ASR-based, Unfreezing strategy=All layers2019.04 | 97.2 | |
| No pre-trainingTraining data size=Full dataset2019.04 | 96.6 | |
| Unfreeze all layersTraining data size=10% subset, Pre-training=ASR-based, Unfreezing strategy=All layers2019.04 | 95.8 | |
| No pre-trainingTraining data size=10% subset2019.04 | 88.9 |