quartznet15x5_LibriSpeech.py training config broken - can not train model

NVIDIA / OpenSeq2Seq

Toolkit for efficient experimentation with Speech Recognition, Text2Speech and NLP

https://nvidia.github.io/OpenSeq2Seq

Apache License 2.0

1.54k stars 369 forks source link

quartznet15x5_LibriSpeech.py training config broken - can not train model #513

Closed cogmeta closed 4 years ago

cogmeta commented 4 years ago

has quartznet15x5_LibriSpeech.py been tested? I am trying to train on model (which has been used before to train jasper models) but validation WER just keep increasing. Something is not obviously right.