Birdnet-go is a real-time sound analyzer that detects bird species. It can take in audio from microphones. I set mine up to stream audio from my security cameras that I already had installed. There’s even a Home Assistant App, if you’re using it.
I live in a more suburban/semi-urban environment, but was really surprised to so many different bird species.
Let this post be the one to ask a question that’s been bothering me (not a coder):
This and similar claims in other software suggest - to me - that the AI is running locally on the raspi, no need to exchange data with sites providing AI APIs or whatever.
I struggle to believe that this is possible, let alone on a raspi.
So, how “local” ar the AIs that help more and more software with various tasks?
Is “local AI” some sort of selling point that is often false or deliberately misleading? “Localwashing”?
edit: thanks for all the answers, this has been enlightening.
I’ve been running birdnetpi on a raspberry pi 3 for years. It runs very well on this modest hardware. I also run birdnet-go in a docker container and use my reolink security cameras for the RSTP audio feed to that instance.
There are all sorts of AI/ML models available that can run on smaller hardware. But they are usually very specific and not general purpose, like the LLMs.
Audio processing by ai can absolutely happen on a pi.
As an example, whisper.cpp is a high performance speech recognition model, and it does a great job. The biggest version of it as a model requires less than 5GB to live fully in memory, the large model at about 1.5billion parameters, and it works great even CPU only.
The one used most often is small, which has about 244million parameters, and needs about a gig of ram and nothing more.
For a pi 5, usually the base or tiny model are used. Small could be used in most scenarios, but tiny and base are able to run real-time.
So it depends for how local - sometimes its all self contained and can do just fine on a pi, sometimes its local in that it needs an LLM endpoint to hit, but that can be running on another system entirely.
It would really depend on the project for how/where/why for a definition of local.
Thanks.
I understand now that this software does not use one of the big general AI providers but a specialised model (for birds esp.?) that runs locally; that I can get behind.
Are there other keywords in software descriptions I can look for that help me make such distinctions?
Aside from “local model”, you might see things like whisper.cpp, llama.cpp, ollama, or “OpenAI compatible endpoint”, or a reference to getting models from HuggingFace (you can think of it like github for llm’s - its more and less, but accurate enough for what we’re talking about here).
whisper.cpp is a c++ implementation of whisper (ASR as mentioned), llama.cpp is a high performance and lightweight tool to support llm/llm inference, ollama is a very user friendly way to do the same, and OpenAI compatible endpoint is just using the protocol for communication, so you can point the client at llama.cpp or similar running on the same machine or another you have access to.
Most things with an extemely tight purpose like this (like listening for some birds singing) will use much leaner models, and may be able to run even from a phone.
Specialized models can be made incredibly efficient. The big tech guys are trying to replace humans which is why their crap takes such huge servers to run. Building local-first/only just means you need to specialize the model and the pipeline to match what you’re doing
Thanks.
I understand now that this software does not use one of the big general AI providers but a specialised model (for birds esp.?) that runs locally; that I can get behind.
Are there other keywords in software descriptions I can look for that help me make such distinctions?
It can run on device on your phone https://birdnet.cornell.edu/
I haven’t verified it, but it local ML models on audio is pretty efficient and should run fine on a raspberry pi.
If you’re speaking of audio analysis ML or AI models, they are quite efficient. As a mediocre musician, who is clinically deaf and cannot hear some frequencies with enough clarity to make a judgement either way, I lean on AI assisted audio mastering. I haven’t dabbled in AI generated music, but with the right DAW and a handful of plugins, some keyboards and instruments, you can do a lot right in the comfort of your home.
Listen, everything that they are doing by scorching the planet and turning the tap water brown, could be achieved on CPU or much smaller GPUs on local machines - it’s a question of focused, tight training. What they’re doing instead is literally attempting to build gigantic models that take in absolutely everything (burning real books in the process because fuck history we got money to make) and attempting to sell the idea that this brute force, one-thing-does-it-all approach, which requires us to give up everything else that is good in the world, is the only way forward, and yes, we must build the roko basilisk now.
It’s a big, big grift.
Didn’t downvote you however, are you saying if I ran an AI model local, that scorching the planet and turning the tap water brown is ok with you, just because it’s local?. Because that’s exactly what we’re are doing when we access the internet. Throw AI out of the scope, and we’re still pumping massive amounts of carbon in to the air, still turning the water brown. There are 8.4 billion people on this planet. Approximately 6.3 billion of them are online. That’s just individual humans. That doesn’t include the hundreds of thousands of data centers that aren’t even connected with AI and all the energy production to make them run.
So, who will be the first to actually practice what you preach and shut down your servers and internet devices? C’mon…make the first step towards a cleaner planet.
This whole internet could run on a tiny fraction of what it current consumes, with no drop in service levels… if the extraction of profit from every single interaction was not prioritized.
That being said, I was addressing someone who refused to believe that local inference was even possible on CPU; this tells me that the efforts to present the current grift as the only and inevitable AI outcome is working, on minds that do not fundamentally grasp what the technology is, nor what it’s doing. I think the tech is great, and highly useful - the way they are implementing has perverse incentives and existential outcomes, and it does not have to be that way, but it is, because capitalism.
But it doesn’t.
But it is.
So, now what are we going to do? Give me real world solutions that will actually fix the issue, which would be implemented by real world corporations and real world data centers. I’m not problem hampered, I’m solution based. So let’s fix it.
It’s a small model that runs on your machine using a TFLite model. AI/ML models come in many different sizes. Some more conducive for running locally than others. There’s more info on the detection pipeline here.
I didn’t write this, just found and self hosted it.
Thanks.
I understand now that this software does not use one of the big general AI providers but a specialised model (for birds esp.?) that runs locally; that I can get behind.
Are there other keywords in software descriptions I can look for that help me make such distinctions?