search

SeanNaren / deepspeech.pytorch

Speech Recognition using DeepSpeech2.
MIT License
2.11k stars 620 forks source link

How to train? #338

Closed ErnstTmp closed 6 years ago

ErnstTmp commented 6 years ago

Hello Sean, thanks a lot for creating this model, this is terrific!!!!!!

Can you comment on how you created the models in Release v1.2 that trains with a WER of 10.2 on Librispeech? Did you use any special traicks on the learning rate selection, ...?

Thanks and kind regards Ernst

ryanleary commented 6 years ago

The models in Release v1.2 are unchanged from the previous release's models and how they were trained is described in release "Pre-trained models V2"