Wav2Letter: an End-to-End ConvNet-based Speech Recognition System

The architecture and flags files are provided for each dataset to reproduce results from the paper (training and decoding steps). Considered benchmarks:

• Librispeech
• WSJ