First, you need to pip install requirements.txt
:
pip [pip3] install -r requirements.txt
You will also need to install torchaudio (and its prerequisites). Note: torchaudio doesn't seem to work with PyTorch 1.1.0 yet, so this repo remains locked to 1.0.0. I will update once this has been fixed.
To run the speech commands dataset, first [download](https://ai.googleblog.co .m/2017/08/launching-speech-commands-dataset.html) it from Google and extract it into your directory of choice. Then you can call
PYTHONPATH=. python [python3] training/commands.py -d PATH/TO/SPEECH/COMMANDS/FOLDER -b BATCH_SIZE -e EPOCHS
Currently, the script makes a random 80/20 split of the data instead of using Google's CSV splits, but this would be easy to modify.