BirdNET-Go is an AI solution for continuous avian monitoring and identification

• 24/7 realtime bird song analysis of soundcard capture, analysis output to log file, SQLite or MySQL
• Utilizes BirdNET AI model trained with more than 6500 bird species
• Local processing, Internet connectivity not required
• Easy to use Web user interface for data visualisation
• BirdWeather.com API integration
• Realtime log file output can be used as overlay in OBS for bird feeder streams etc.
• Minimal runtime dependencies, BirdNET Tensorflow Lite model is embedded in compiled binary
• Runs on Windows, Linux and macOS
• Low resource usage, works on Raspberry Pi 3 and equivalent 64-bit single board computers

Ready to run binaries can be found from releases section https://github.com/tphakala/BirdNET-Go/releases/ Archives also contains libtensorflowlite_c library.

docker run -ti \
  -p 8080:8080 \
  --env ALSA_CARD=<index/name>
  --device /dev/snd \
  -v /path/to/config:/config \
  -v /path/to/data:/data \
  ghcr.io/tphakala/birdnet-go:latest

Within the BirdNET-Go container, knowledge of the designated microphone is absent. Consequently, it is necessary to specify the appropriate ALSA_CARD environment variable. Determining the correct value for this variable involves the following steps on the host computer:

1. Open a terminal and execute the command arecord -l to list all available capture devices.

> arecord -l
**** List of CAPTURE Hardware Devices ****
card 0: PCH [Generic Analog], device 0: Analog [Analog]
  Subdevices: 1/1
  Subdevice #0: subdevice #0
card 0: PCH [Generic Analog], device 2: Alt Analog [Alt Analog]
  Subdevices: 1/1
  Subdevice #0: subdevice #0
card 1: Microphone [USB Microphone], device 0: USB Audio [USB Audio]
  Subdevices: 1/1
  Subdevice #0: subdevice #0

2. Identify the desired capture device. In the example above, cards 0 and 1 are available.
3. Specify the ALSA_CARD value when running the BirdNET-Go container. For instance, if the USB Microphone device is chosen, set ALSA_CARD to either ALSA_CARD=1 or ALSA_CARD=Microphone.

Download precompiled TensorFlow Lite C library for Linux from https://github.com/tphakala/tflite_c/releases/tag/v2.14.0

Copy libtensorflowlite_c.so to /usr/local/lib and run ldconfig

sudo cp libtensorflowlite_c.so /usr/local/lib
sudo ldconfig

Clone tensorflow repository, this is required for header files to be present while compiling with CGO

mkdir ~/src
cd ~/src
git clone https://github.com/tensorflow/tensorflow.git

Checkout TensorFlow v2.14.0 release

cd tensorflow
git checkout tags/v2.14.0

Clone BirdNET-Go repository

git clone https://github.com/tphakala/BirdNET-Go.git

Build BirdNET-Go by make, compiled binary will be placed in go-birdnet/bin directory

cd BirdNET-Go
make

Windows build is cross compiled on Linux, for this you need MinGW-w64 on your build system

  sudo apt install mingw-w64-tools gcc-mingw-w64-x86-64 gcc-mingw-w64-i686

Download precompiled TensorFlow Lite C library for Windows from https://github.com/tphakala/tflite_c/releases/tag/v2.14.0

Copy libtensorflowlite_c.dll to /usr/x86_64-w64-mingw32/lib/

sudo cp libtensorflowlite_c.dll /usr/x86_64-w64-mingw32/lib/

Clone tensorflow repository, this is required for header files to be present while compiling with CGO

mkdir ~/src
cd ~/src
git clone https://github.com/tensorflow/tensorflow.git

Clone BirdNET-Go repository

git clone https://github.com/tphakala/BirdNET-Go.git

Build BirdNET-Go by running make windows

cd BirdNET-Go
make windows

Windows executable is in bin/birdnet.exe, copy this and libtensorflowlite_c.so to your Windows system, library file must be in PATH for birdnet.exe to run properly.

Yes it is correct that you need libtensorflowlite_c.dll in /usr/x86_64-w64-mingw32/lib/ for compile process, and on Windows you need libtensorflowlite_c.so for runtime. This sounds backwards but this is how it works.

BirdNET-Go CLI

Usage:
  birdnet [command]

Available Commands:
  authors     Print the list of authors
  completion  Generate the autocompletion script for the specified shell
  directory   Analyze all *.wav files in a directory
  file        Analyze an audio file
  help        Help about any command
  license     Print the license of Go-BirdNET
  realtime    Analyze audio in realtime mode

Flags:
      --debug               Enable debug output
  -h, --help                help for birdnet
      --locale string       Set the locale for labels. Accepts full name or 2-letter code. (default "finnish")
      --overlap float       Overlap value between 0.0 and 2.9
      --sensitivity float   Sigmoid sensitivity value between 0.0 and 1.5 (default 1)
      --threshold float     Confidency threshold for detections, value between 0.1 to 1.0 (default 0.8)

Use "birdnet [command] --help" for more information about a command.

There is more detailed usage documentation at wiki https://github.com/tphakala/birdnet-go/wiki

Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International

Tomi P. Hakala

BirdNET AI model by the K. Lisa Yang Center for Conservation Bioacoustics at the Cornell Lab of Ornithology in collaboration with Chemnitz University of Technology. Stefan Kahl, Connor Wood, Maximilian Eibl, Holger Klinck.

BirdNET label translations by Patrick Levin for BirdNET-Pi project by Patrick McGuire.