Streamable KWS with a minimal number of model parameters.
(As of now it achieves ~94.3% test accuracy on the Speechcommands v2 Dataset, 35 Classes. phi_gru.yaml config The model should weight around 43k parameters.)
To run the script:
python3 train.py --config-name=name-of-config.yaml