Phone recognition on TIMIT (dev)
28.5Frame Error RateBidirectional LSTM
Evaluation results
| Method | Links | |
|---|---|---|
| Bidirectional LSTMArchitecture=250 cells, Layers=52015.04 | 28.5 | |
| Bidirectional iRNNArchitecture=500 neurons, Layers=52015.04 | 28.9 | |
| Bidirectional LSTMArchitecture=250 cells, Layers=22015.04 | 29.6 | |
| Bidirectional RNNArchitecture=500 neurons, Layers=22015.04 | 31.5 | |
| Bidirectional iRNNArchitecture=500 neurons, Layers=22015.04 | 31.9 | |
| iRNNArchitecture=500 neurons, Layers=52015.04 | 33 | |
| Bidirectional RNNArchitecture=500 neurons, Layers=52015.04 | 33.9 | |
| iRNNArchitecture=500 neurons, Layers=22015.04 | 34.3 | |
| LSTMArchitecture=250 cells, Layers=22015.04 | 34.5 | |
| RNNArchitecture=500 neurons, Layers=22015.04 | 35 | |
| LSTMArchitecture=250 cells, Layers=52015.04 | 35 | |
| RNNArchitecture=500 neurons, Layers=52015.04 | 35.6 |