TF-Speech-Recognition

Experimental code for training CLDNN models using tensorflow.

Model Architecture

• 2 Conv layers with larger filter size + 5 residual block (15 conv layers) + 5 residual LSTM

Results on LibriSpeech

• Trained on LibriSpeech-train-clean-100

Validation-Set	WER	Link
Librispeech-test-clean	NA	NA

Training logs

• About 30 epochs
• The final loss approaches 45-50.