/OrcaML

Cetacean detection from hydrophone audio via image classification on spectrograms.

Primary LanguageJupyter NotebookMIT LicenseMIT

OrcaML alt text

  1. pip install requirements.txt

Instructions for Testing:

  1. Download medium_orca.mdl from here.
  • Put medium_orca.mdl in models/
  • Run:
    python3 process_wavs.py --path /path/to/data/
    on folder of 2 second long snippets of whale/background sounds. This will create a folder /path/to/data/proccessed/ with resampled wavs and spectrograms in .npy format ready to be input into the model.
  • Run Train Model and Test SKRW cells in OrcaML.ipynb, making sure to update paths to point to the processed/ folder from step 3.

Instructions for Training:

  1. Process data as in step 3. of Instructions for Testing.
  2. Place data into data/ structured as shown in the first cell of OrcaML.ipynb
  3. Uncomment training in Train Model cells of OrcaML.ipynb and run.