Release of online RNN-T baseline

[swpd]

Hi,

Thanks for your great work of the online speech recognition. I'm also insterested in the RNN-T baseline mentioned in the paper and would like to compare it with TDS model on my own training data. According to the paper Section 4.2,

All our experiments are run using wav2letter++ framework [2] with 64 GPUs for each experiment.

Can you please consider releasing the RNN-T baseline? (RNN-T train/decode recipes, RNN-T critcriterion, etc.)

Thanks!

[vineelpratap]

Hi, For LC-BLSTM + RNN-T baseline, we got the trained model from the authors of https://arxiv.org/abs/1911.01629 and used it to compute the experimental results. It is not trained in Wav2letter++ framework.