I have some 25,000 hours of acoustic recordings to process via BirdNET analyzer, most of it in 15 second chunks. I ran an initial ~4,000 hours, which took a few weeks running in the background on my laptop, and I suspect contributed to my laptop dying. It was old and not up to the task, considering the noise the fan was making. I now have a new laptop with a better GPU and CPU, which will hopefully help, but I don't want to burn this one out too!
Has anyone run Birdnet on high volumes of data like this? Any advice?
In graphics settings (I'm on windows 11), should I set my GPU preference to high performance rather than 'let windows decide'?
17 May 2025 10:44am
I haven't tried BirdNET analyzer, but with regards to running any bigdata/ML processing, my advice would be to look at something like Google Colab instead of your own laptop.
Hope this helps.
19 May 2025 12:22am
Would that be able to process locally stored acoustic data?
One of the great things about birdnet analyzer is that it is local - it doesn't require uploading terabytes of data into the cloud, which would be expensive, take forever, and likely have some transfer errors in areas with poor internet connection (like the tropics where I do my research).
30 May 2025 2:59pm
Hi Hubert, I think the key will be to ensure the model is making use of your GPU. In general, BirdNET should theoretically support GPU usage but I have never been able to get it working. You can use a tool such as nvidia-smi to ensure that the GPU is being used.
My recommendation (as the developer behind the bioacoustics model zoo) would be to use WSL 2 (windows subsystem for linux) and use one of the alternatives to BirdNET available in the the bioacoustics model zoo (Perch, BirdSet EfficientNet, BirdSet ConvNext, HawkEars). All of these alternatives provide GPU support. The also might work perfectly fine on Windows without WSL, but you're more likely to encounter issues.
GitHub - kitzeslab/bioacoustics-model-zoo: Pre-trained models for bioacoustic classification tasks · GitHub
Pre-trained models for bioacoustic classification tasks - kitzeslab/bioacoustics-model-zoo
this will require a bit of python environment set up and coding, but WSL + python + bioacoustics model zoo will ensure you can use your GPU to speed up inference. At a high level, you would need to
(1) install WSL if you don't have it yet
(2) in WSL, install miniconda and create a conda environment
(3) install the bioacoustics model zoo in the conda environment
(4) run a python script using a model zoo model .predict(audio_files) function. You'll want to use a large batch_size eg 256 and probably num_workers=4 or so if you have multiple CPU cores to use
See docs in the link below and at opensoundscape.org for details, and feel free to reach out if you need help
Cheers,
Sam
GitHub - kitzeslab/bioacoustics-model-zoo: Pre-trained models for bioacoustic classification tasks · GitHub
Pre-trained models for bioacoustic classification tasks - kitzeslab/bioacoustics-model-zoo
25 November 2025 11:16am
I wanted to expand on the OPs query in relation to Apple hardware.
I'm trying to work out the fastest and best way to run BirdNet on behalf of a charity (we are a commercial company but provide probono support to a local rewilding charity) on an enormous dataset (1.6TB for three seperate datasets - purely becasue the whole 3.9TB dataset is too big to process in one go).
I have a 2024 Mac Mini M4Pro (14 core CPU; 20 core GPU; 48GB Memory; 1TB onboard SSD; 4TB Thunderbolt 5 SSD).
Is there any advice on how best to approach this task other than subsampling the datasets (which will still be a large dataset irregardless).
Any help at all will be appreciated :-)
Ajay Bhandari