BIRDNETFS: BirdNET compatible with fsspec

Introduction

BirdNET is great for analyzing files stored locally but we sometimes want to analyze files that are stored remotely. This repository aims to close that gap by using a modified version of BirdNET-Analyzer/analyze.py that leverages fsspec, a python library that help open files that are stored remotely.

How to use it?

This repository is made so that we can pull the changes made to the BirdNET repository without any conflict.

1- Clone this repository and install the dependancies

# Clone this repository:
git clone https://github.com/NINAnor/birdnetfs.git
cd birdnetfs
pip install requirements.txt

2- Clone the BirdNET repository:

git clone https://github.com/kahst/BirdNET-Analyzer.git
mv BirdNET-Analyzer birdnetsrc

3- Analyze

There is two options.

First you can analyze a file of your chosing using:

export PYTHONPATH="${PYTHONPATH}:./birdnetsrc"
python analyse.py filecache::ssh://$USER:$PASSWORD@HOST:/PATH/TO/AUDIO/FILE1.mp3

Or analyze multiple files in parallel (using GNU parallel):

In files_to_analyze.csv list the files that you want to analyze

Then run:

sudo apt-get install parallel
time systemd-run --scope --user --property=CPUWeight=1 -- sh -c './analyse.sh'

Analyzing the files will return Birdnet.selection.table.txt files in the OUTPUT_PATH_BIRDNET.

Extract the detections

1- Update the config_connection.yaml

2- Build a .parquet database using:

python3 src/parse_results.py

⭐ Note that the database will contain $NUM_SEGMENT at $THRESHOLD. Both parameters saved in config_connection.yaml.

3- Extract the detections!

python3 src/extract.py