README.md

avian-bioacoustics-oesf

Avian biodiversity and vocalization behavior in Washington's Olympic Experimental State Forest

Recommendations

• Visual Studio Code with Microsoft extensions Python, Python Debugger, and Jupyter
• Install dependencies to a virtual python environment (.venv) that is used exclusively for this project. Create the environment manually, or via VS Code.

Prerequisites

• Python 3.9+ 64-bit (3.10 recommended, ensure "Add path to environment variables" is checked during install)

Package dependencies

From a terminal shell within the virtual environment, navigate to the root directory of this repository (.../avian-bioacoustics-oesf), and run:

git submodule update --init --recursive
pip install --upgrade pip
pip install -r requirements.txt

This should install all dependencies you need to run, for example, analyze_file.py. For further reference, see setup instructions for BirdNET-Analyzer and birdnetlib.

Optional: Sound separation dependencies

This repository has a fork of bird_mixit as a submodule, which is used to perform sound separation.

1. Follow steps to install gsutil
2. Install bird_mixit model checkpoints via a terminal shell (see below)

gsutil -m cp -r gs://gresearch/sound_separation/bird_mixit_model_checkpoints data/models/

Contents

• analyses – ecological analyses and models (e.g. species accumulation curves)
• annotation – extracting samples for annotation and clustering
• classification – run classifier on audio/directory, evaluate performance
• data – data associated with the study area and project

Training and performance evaluation pipeline

1. Manually annotate training examples with Raven Pro
2. Run training_extract_audio_examples.py to extract audio examples for training
   • Manually add any additonal class examples (e.g. "Background") to the audio subdirectory
3. Run training_assemble_datasets.py to assemble datasets for training
4. Train custom classifiers with TODO

cd src/submodules/BirdNET-Analyzer/
python3 train_custom.py --i /Users/giojacuzzi/repos/avian-bioacoustics-oesf/data/models/custom/custom_S1_N2_A0/training_files.csv --o /Users/giojacuzzi/repos/avian-bioacoustics-oesf/data/models/custom/custom_S1_N2_A0/custom_S1_N2_A0.tflite --no-autotune

4. Return to top directory with cd ../../../, then run test_compare_validation_performance.py