To replicate training process clone this repository, install the requirements.txt, update config.yaml and run command from the command line:
python3 -m src.train
Note: following script implies logging via Weights&Biases, please run: wandb loging
to log into your account.
As we train model using cross-validation this repository does not consist a separate script to compute accuracy on validation folder, insted we can run evaluate.py script to predict class for the audio file by running:
python3 -m src.evaluaton.evaluate -p <path to the .wav file> -w <path to the model weights>
An example of running evaluate.py script can be found in the notebook, feel free to open it in Google Colab, to listen and observe model predictions for 4 random samples from urbansound dataset.