training.rst

Training

After dummies_tutorial_data_preparation, we can start training.

The command to start the training is quite simple:

cd /tmp/icefall
export PYTHONPATH=/tmp/icefall:$PYTHONPATH
cd egs/yesno/ASR

# We use CPU for training by setting the following environment variable
export CUDA_VISIBLE_DEVICES=""

./tdnn/train.py

That's it!

You can find the training logs below:

./code/train-yesno.txt

For the more curious

./tdnn/train.py --help

will print the usage information about ./tdnn/train.py. For instance, you can specify the number of epochs to train and the location to save the training results.

The training text logs are saved in tdnn/exp/log while the tensorboard logs are in tdnn/exp/tensorboard.