README.md

Scaling Up Online Speech Recognition Using ConvNets

We are releasing model limited future context of 500ms for the streaming pipeline. This model is trained on supervised Librispeech (1k hours) and unsupervised LibriVox (57k hours) data.

Details on architecture, flags files and reproducing steps one can find in the librispeech directory.