Active Noise Cancellation of Drone Propeller Noise using machine learning

Code instruction

To run the code, make sure to have the correct dependencies:

pip install -r requirements.txt

Training

To train the GCRN model, change line16 in model.py to choice = 'GCRN', then run python model.py

To train the LSTM model, change line16 in model.py to choice = 'simple', then run python model.py

After training, you will get a .png file and a .txt of the loss, a .wav file, which is the anti-noise signal, and a .pt file of the trained model.

Testing

If you want to test a specific .pt file, you can use python test_model.py. Remeber to change the path to the .pt file and the model type (please specify in choice).

Real-time set up

You can run a real-time experiment using python test_sd.py. You can specify your model in line 11.

Audio Results

The noise sample used for testing is test_noise_1.wav.

The anti-noise signals generated are recon_signal_GCRN_100.wav and recon_signal_simple_100.wav corresponding to the 2 models.

The resulting sound is in Result_sound is the combination of the noise and anti-noise signal. The combination here means use different speaker to output the noise.

Acknowledgement

The network.py is modified from https://github.com/JupiterEthan/GCRN-complex/tree/master.