DeepSpeech model training accuracy

I can’t see the training accuracy for https://github.com/mozilla/DeepSpeech. All I see is the loss.

They have it on the release notes usually.

No, I am saying when I train the model…

Train more: test set will be executed at the end.

1 Like

I ran this script:

!./DeepSpeech.py --train_files "/gdrive/My Drive/data_speech_commands_v0.02/train.csv" --test_files "/gdrive/My Drive/data_speech_commands_v0.02/test.csv" --export_dir "/gdrive/My Drive/Data_op" --checkpoint_dir "/content/chkpoint"

But I kept seeing the following output:

Epoch 35 | Training | Elapsed Time: 0:04:18 | Steps: 3276 | Loss: 18.129789
Epoch 36 | Training | Elapsed Time: 0:03:02 | Steps: 2300 | Loss: 17.967608

But it crashed before I could see accuracy. Is it supposed to output accuracy after training?

As said, and documented, test set is run after full training.

A friend of mine ran it fully and this is the output he got:

Please follow the forum guidelines and avoid sharing such material as screenshots.

I am sorry. But can you please tell me what could be going wrong and why the accuracy is not listed at the end of the training?

It is, see the “WER”.

So does that mean there is 100% accuracy?

We are happy to help, but you have given us next to zero information. Try training a really small dataset completely without error and start from there. If you need more:

You need to do your homework and search for Word Error Rate.