Continuing training from a frozen graph

[JRMeyer]

>>> krawat
[January 25, 2018, 12:03pm]

Is there some indicator in logs that the model was loaded
(https://github.com/mozilla/DeepSpeech#continuing-training-from-a-frozen-graph)?
I used pre-trained model to boost learning but there is only a slight
difference in results, so I wonder if I've done everything right.

python -u DeepSpeech.py slash --n_hidden 2048
slash --initialize_from_frozen_model output_graph.pb slash --train_files train.csv
slash --dev_files valid.csv slash --test_files test.csv slash --train_batch_size 64
slash --dev_batch_size 64 slash --test_batch_size 64 slash --epoch 1 slash --validation_step
1 slash --dropout_rate 0.30 slash --default_stddev 0.046875 slash --learning_rate
0.0001 slash --checkpoint_dir checkpoint_transfer/ slash --summary_secs 600
slash --summary_dir tensorboard_transfer/

[This is an archived TTS discussion thread from discourse.mozilla.org/t/continuing-training-from-a-frozen-graph]

[JRMeyer]

>>> reuben
[January 25, 2018, 12:49pm]



> Is there some indicator in logs that the model was loaded

Yes. At the beginning of training we log slash 'Initializing from frozen
model: slash ' followed by the path of the frozen model.

[Archived Post]

[JRMeyer]

>>> elpimous_robot
[January 25, 2018, 8:06pm]



> I wonder if I've done everything right.

Yes, it seems, but how do you call your new model, for inferences ??

you called the existing model to adapt :
slash --initialize_from_frozen_model slash
did you save the new, modified one ? : slash --export_dir

test your new model.

> I personnaly tested with my homemade limited model, adapting my model
> with my 8 years old son (first time). slash
> Inferences, not perfect were good. slash
> Without adaptation, impossible to read a correct inference about my
> child voice. slash
> So, frozen_model adaptation successed ! slash
> hope to help

[Archived Post]