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
python -m venv .venv
source .venv/bin/activate
pip install requirements.txt

2- Clone the BirdNET repository:

birdnetfs attempts to reuse BirdNET functions as much as possible.

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

3- Analyze

First, you need to change the BirdNET config_file (NOT the config_connection.yaml) located in src/config.py. More particularly you may want to change the OUTPUT_PATH parameter which is the path where the output files will be created.

There is two options.

• First you can analyze a file of your chosing using:

export PYTHONPATH="${PYTHONPATH}:./src/birdnetsrc:src:birdnetsrc"
python analyse.py filecache::ssh://$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 sample.parquet database using:

python3 src/parse_results.py

This parquet database is a database containing ALL the results from BirdNET.

3- Build a sampled_segments.parquet

python3 src/global_sampler.py

⭐ Note that this parquet file will contain $NUM_SEGMENT random segments with the $THRESHOLD indicated in the config_connection.yaml.

4- Extract the detections!

./extract.sh

Format for annotation

To annotate the extracted segments, we create a csv file per species. The csv list the segments to annotate and add a few columns for the annotator to fill. To help create the csv files, the repository contains a bash script to_annotation_sheet.sh.

Be sure to change the input and output in the script L4 and L5:

BASE_DIR= ./extracted_segments # Folder where all the segments are stored 
OUTPUT_DIR= ./csv # Folder that receives the csv files

Then run the script:

./to_annotation_sheet.sh