Deep Learning for Speech Recognition
The model is using Connectionist Temporal Classification implementation from Baidu Search.
install-deepspeech-torch.sh is a step by step walkthrough to install the required componants for training the deepspeech.torch model from Sean Naren (https://github.com/SeanNaren/deepspeech.torch).
A first run on an AWS EC2 g2.8xlarge (2 CNNs, 7 RNNs with 1760 units) has been performed on a 50 minutes clean dataset.
UPDATE: Trained the model on LibriSpeech dataset 1000h of speech: 12% of WER (test-clean for the validation set). On 2xK80 with batch size 64 - almost 10 days on run.