Train the DNN

The parameters of the DNN will be learned using stochastic gradient descent.

Edit your config file to turn training on

NORMLAB  : False
MAKECMP  : False
NORMCMP  : False
TRAINDNN : True
DNNGEN   : False
GENWAV   : False
CALMCD   : False

and change a few values to make training very fast, just for testing

[Architecture]

training_epochs  : 5
[Data]

train_file_number: 10
valid_file_number: 5
test_file_number : 5

and run this step

$ python /Volumes/Network/courses/ss/dnn/dnn_tts/run_dnn.py feed_forward_dnn_WORLD.conf 

Ignore any warning from then about “UserWarning: The Param class is deprecated…”

You can also obtain plots from the training process by adding these to your configuration file (you may need to create the [Utility] section):

[Paths]
plot_dir : log

[Utility]
plot : True