This script will download bird audio recordings from https://www.xeno-canto.org, split them into 8 second files, and create training and test datasets of the desired size.
A noise class will be added using background noise downloaded from the following sources:
- https://github.com/microsoft/MS-SNSD
- https://ai.googleblog.com/2017/08/launching-speech-commands-dataset.html
Training sets will be balanced, so each class will have the same number of samples. Test sets include all available remaining samples.
To download the default audio dataset (oak titmouse and mourning dove), run the script prepare_data.py using python 3. You will need to install any dependencies that are missing.
Audio for different birds can be downloaded by editing the init and if __name__ == "__main__": sections of the file to specify the desired species.
The script in upload_all.sh provides an example of how to upload the data to Edge Impulse for the default dataset. You will need to modify this if you add different birds.
This is a rough script provided for experimentation only.