The .ipynb files contain our four models:
- Baseline ANN
- 1D CNN
- CNN with spectrogram input
- Transfer Learning with ResNet
We did most of our training in Google Colab to access their GPUs.
Our ANN and 1D CNN use the chromagram data of the audio files as input,
while the CNN-spectrogram and Transfer Learning ResNet use spectrogram data (frequency/time).
Does it make a difference?
If you're interested in our results, take a look!