Silent Speech Recognition on EMG open-vocabulary 2020 (test)
42.2WERTransduction model
Evaluation Results
| Method | Links | |
|---|---|---|
| Transduction modelArchitecture=Transformer, Feature Extraction=Convolution, Loss Components=Phoneme loss included2021.06 | 42.2 | |
| Transduction modelArchitecture=Transformer, Feature Extraction=Hand-designed features, Ablation=Replace convolution features with hand-designed features2021.06 | 45.2 | |
| Transduction modelArchitecture=LSTM, Feature Extraction=Convolution, Ablation=Replace Transformer with LSTM2021.06 | 46 | |
| Transduction modelArchitecture=Transformer, Feature Extraction=Convolution, Loss Components=Remove phoneme loss, Ablation=Remove phoneme loss2021.06 | 51.7 | |
| Gaddy and Klein (2020)Architecture=LSTM (3 bidirectional layers, 1024 dimensions), Feature Extraction=Hand-designed features2021.06 | 68 |