Experimental code for training CLDNN models using tensorflow.
- 2 Conv layers with larger filter size + 5 residual block (15 conv layers) + 5 residual LSTM
- Trained on LibriSpeech-train-clean-100
| Validation-Set | WER | Link |
|---|---|---|
| Librispeech-test-clean | NA | NA |
- About 30 epochs
- The final loss approaches 45-50.
