vrenkens / nabu

Code for end-to-end ASR with neural networks, build with TensorFlow
MIT License
108 stars 43 forks source link

Training loss over 400? #34

Closed rzcwade closed 6 years ago

rzcwade commented 6 years ago

Hi @vrenkens ,

I got your LAS + WSJ model training working but I got training loss over 400 to start with. What's your results like?

Thanks.

vrenkens commented 6 years ago

This is probably because your using alignments, check other issue for a response :)

rzcwade commented 6 years ago

@vrenkens , thanks for your response. I think it's because it uses the sum of cross entropy as loss function.