Are there any LLMs being widely used for audio classification? I know VLMs are being used a lot in image stuff.
It always seems kind of silly to me to throw everything at an LLM. I know they’re huge and can automatically handle a huge number of tasks but something in me finds it wasteful when we could be creating easily trainable, cheap to run bespoke models for a lot of stuff
e-ink is so much fun, especially when combined with ESP32 or BTLE boards. I currently have 4 around my house displaying book quotes that I've highlighted in KOReader and they bring me joy each time I see them. They are simple and do one thing very well.
I just got and setup a BTLE e-ink driver the other day and according to my calculations it will last _years_ on a single charge (2000mAh) even with multiple refreshes per day (crazy compared to the Wifi ones). Year+ lifetime completely changes the calculus for where to put these IMHO.
I'm considering designing a multiple-screen "art" piece to hang on my wall and playing around with various ideas (on both content and design).
LLMs + 3D Printer + e-ink is my new favorite hobby.
Here are the parts I bought, I'm still getting it all setup (I have my pictures displaying, working on designing the 3D printed case now, and testing it all) but according to OpenDisplay's calculator [0] I can expect 3.4 years if I refresh the screen once every 4 hours (and currently I'm doing closer to once a day to mimic my ESP32s which need deep sleep and once-a-day refresh to get even month/months of battery on a charge).
The image rendered to the screen is coming from a little service I wrote that renders a book quote it pulls from BookOrbit which syncs with KOReader (annotations, progress, etc). My plan is to set up one of these one each bookshelf/series rotating quotes I highlighted from the series.
Here are 2 ESP32-based screens I have, the one on the left is a an all-in-one [1] and the one on the right I printed the case from the TRMNL DIY kit from SeeedStudio [2]: https://cs.joshstrange.com/knk4Ns2h
Right now only the BTLE one is running OpenDisplay (the other two are semi-managed by HomeAssistant ESPHome) and all the devices are in HA for controlling them, though the ESP32's all deep-sleep for saving battery so I can can't "push" stuff to them like I can the BTLE one.
If I want a wifi e-ink connected display - curious what would you recommend.
I want an e-ink display - that I can use as a sign/display board with wifi connectivity and remote administration.
If you want a simple e-ink wifi screen I'd point you at the reTerminal e1001 [0] if that size works for you. It's the easiest to get started with. You can either keep it on power or use the deep sleep to get more from the battery. It really comes down to how often you want to update it.
The BTLE boards have just crazy higher battery time with no need for deep sleep which is why I'm moving in that direction.
~7.5" is the size I've been buying so far, the larger e-ink screens get pricey but I can't wait for a calendar-sized one to come down in price such that I can frame it and hang it on the wall. Lastly I'd avoid the color e-ink screens unless you know you need color, won't need to refresh it often, and it's not in your eye line. Color refreshes take way longer (~30s in my experience for a 5-6 color screen) and the flashing it has to do is obnoxious. Monocolor displays take <2s normally and a single flash. I got a color screen first because "why not?" and I got color, but the refresh was too annoying so now it refreshes once a day in a place that I'm not when it refreshes so I don't have to see it.
My mistake for calling it "copy", should have said fork maybe (?)
They are awful close to still require a reference to the original project imo. They are achieving the same output (identify birds and display them in a frame display all hosted in a raspberry pi) with that twist to how to generate the art.
Author here: I'll take any feedback and critisism. I haven't based any of my work directly on his project, but will be referencing it as a similar one in the readme, as well as e.g. https://github.com/veteranbv/inky-bird-frame.
From the link: "Half the point of this project is showing off some amazing public-domain natural-history illustrations. Over 800 cut-outs covering more than 400 species, every one taken from a real plate and hand-curated for this project (no art is AI-generated, though some has been retouched with AI)."
I agree I would expect a reference. Another difference between this an AvianVisitors is the image sources. AvianVisitors is all AI generated with a consistent prompt, but OP's is using chiefly public domain images with retouching by AI as needed.
I understand OP is continuing the work and improving it as he sees fit, but that in my mind is a fork and should aknowledge the initial work that sparked it.
Maybe i'm wrong tho and this is completely parallel work! Waiting for OPs reply
Hi, yeah I'm not basing any of my coding or feature-work on that project. My biggest source of inspiration is actually just a paper poster I have hanging on my wall https://www.axelthorenfeldt.com/news/wwf-verdens-naturfonds-.... I wanted to make a version of that, showing the actual birds in my garden.
A fork implies that it builds upon code from another repository. I wouldn't use the word fork to mean "inspired by". I think it takes away from the effort that the author has put in. Anyone can create a fork of a project with the click of a button.
Original author here. Fair point, this project was one of my inspirations, but I chose to take a different approach. This uses public domain natural history art and builds on top of BirdNET-Go which is a popular way of self-hosting bird-detections. (This is really a BirdNET-Go companion, as it cannot run without).
That's fair, and I totally get wanting to give it your own spin.
I still think it would be good to acknowledge the original project that sparked the idea, especially now that you've mentioned it was one of the inspirations.
Definitely keep going with yours and evolve it further. That's how good projects grow. But for a healthy open source community, I think we should make a point of crediting the people whose work inspires us. Otherwise, over time, it can discourage people from sharing their projects publicly in the first place.
Contrary to the Great Man version of how we mostly talk about history, scientific discoveries, technological breakthroughs and such have historically often occurred multiple times in multiple places often in close proximity time-wise but otherwise completely unrelated to each other. This of course makes sense if you consider that history seems to be primarily driven by material conditions (e.g. steam engines were known to Ancient Greece but their usefulness for industry simply hadn't occurred to them because they lacked the prerequisites to make them useful and neither economic efficiency nor access to cheap labor were significant enough concerns - plus there was no patent system of course).
I'll leave it up to you to decide whether one should expect the demographic of this website to be especially invested in the Great Man narrative but trying to convince them that two people can independently have similar-sounding ideas involving similar-but-different technology is probably an uphill battle. Just be glad you're not composing music for a living because there's literally an entire industry built around the idea that it's impossible for two people to come up with similar melodies without knowingly or unknowingly copying eachother and that it's very important that one of them (or rather the company they had to sign a deal with in order to make any money off their music at all) 100% owns that exact melody and should be paid by anyone who wants to use it.
Next time you have an idea you should just file a patent tbh - the insistence on prior art invalidating the work seems to be quite similar but at least if you're granted a patent you can shake down "copycats" for money.
I haven't tried with BirdNet, but Merlin doesn't seem to ever false-positive on chipmunks or squirrels (there's been a suggestion that it should explicitly recognize them, and maybe some frogs too, since they're "loud thing in the same register" and it frustrates people that haven't figured it out yet...)
I wish E-ink displays were a bit more affordable. The 13 inch one is £229.50.
Does anyone know why they are still so expensive, at least at large sizes?
You have to distinguish the fake color e-ink that are used in mass consumer products which are actually black&white e-ink plus an LCD layer and are expensive, and the real color e-ink where each micro capsule embeds 3 to 5 different inks of different colors.
The latest are pretty complex devices, very slow to refresh (hence why they are not used in your e-reader) and also very expensive. But the rendering is way much nicer because it’s real ink that is used to render the colors.
I just happen to have an extra Pi and an e-ink display sitting around and I LOVE this idea. Going to try it on my screened in porch - we typically get 10-20 Merlin detections on a decent day so this could be fun!
What would be cool is if you can put AI to identify individual birds. They all have different calls/voices, though largely indistinguishable to human ears. And you can name them and see when they come for a visit.
On a side note. How do people deal with all the ugly mess birds cause in such settings? I'd love to have such thing but seeing all the white stuff on glass door does not really feel fun to me overall.
What do you mean specifically? These birdsong analyzers are sensitive enough that they do not rely on setting up a feeder. It’s obviously dependent on where you live, but in the Nordics, living outside city centres is good enough for both background noise and birdsong from nearby trees/parks/forests. Privacy matters aside, setting a mic up on a balcony or window is enough. Another solution is feeding data from a cheap security camera with a mic through wifi. This way the camera can stay outside in different weather, it’s not super close to you actual living area and splitting audio from the RTSP stream should be relatively easy to be analyzed by BirdNET.
Well I specifically meant having a bird feeder on a window or glass door for a chance to watch them live closely. That's why I put it as a side note ;)
You gave me few ideas for the analyzer setup too - thanks!
Sounds like you've got birds crashing into the door? That's often something that can be improved (with stickers or hanging decorations) that make the glass more perceptible to the birds. (Another thing that sometimes helps is having bushes or shrubs, or even potted plants, in sight of the feeder - most small birds will use a "staging area" and fly somewhere nearby first, "check out the scene", and then fly to the feeder from there; it doesn't really control the activity, but it adds some options.)
Thanks for sharing. Didn’t know about this project. Have you used it yourself? I’ve only experimented with Scrypted for fun, but Frigate sounds like I could operate birdsong analysis without a separate BirdNET container.
I can't help wonder what this would look like for humans. Imagine, set one of these up at a party and it has the means to identify anyone on the invite list. When it hears the person it displays a photo, brief bio, and relationship status.
Author here. A mic listens to the sounds in my garden, BirdNET-Go (an open-source BirdNET classifier) detects birds by sound, and the e-ink frame draws a real 1800s natural-history illustration of each one as it hears them. Here’s a live web-demo from my garden in Bergen, Norway: https://fugleramme.arnegiacomo.dev
A few things that might be interesting:
- The screen only redraws when the set of birds changes, and dithers the collage down to six colours for the e-ink display.
- Bigger birds sit toward the centre, scaled by real body mass.
- 800+ cutouts across 400+ species, each cut from a real public domain plate. All art is historic and human-made. Coverage is currently best for the Nordics, Britain and Germany (but other parts of the world are in the works).
- Runs fully local on a Raspberry Pi or your homelab (yes, even the classifier runs great on a RPI)
The e-ink panel is optional and it can run web-only in a container against a BirdNET-Go you already have.
If I were to make one of these as a Christmas present for somebody in Texas, and I wanted to contribute work toward good coverage for the area, how big of a time commitment would that be? It sounds like the kind of hobby work I could get into for a few months.
Less than you'd think, because you likely wouldn't need full "Texas coverage", you mostly need the birds that actually show up. BirdNET filters by location, you'll likely need a couple of dozen. Over the summer I've had about 40 visitors where I live.
Right now the coverage of North America is relatively sparse, but the interest has been high and I'm hoping for some contributions.
Edit: For North America the most obvious source is Audubon's Birds of America, which is public domain and should provide a lot of coverage. This collection includes some beautiful plates.
Do you have your image source available somewhere? I'm not going to build this same thing, but I'd love to add some prints of birds to my shelves and, not knowing much about birds, it would help to have a set of pictures I could look through and go, "Ooh! This one!" and learn about them. It'd be a good jumping on point!
I'm actually currently working on implementing Iberian species, and added some yesterday. There's another user also contributing with some beautiful Spanish birds. Feel free to contribute if you have some cool visitors in your garden!
We could also leave fake images aside, go outside and try to watch the birds ourselves. Maybe even let the camera inside, just looking with our own eyes. Real nature. No intermediate.
> Half the point of this project is showing off some amazing public-domain natural-history illustrations. Over 800 cut-outs covering more than 400 species, every one taken from a real plate and hand-curated for this project (no art is AI-generated, though some has been retouched with AI).
You can in fact both have an interesting bird-based art display in your home and also go outside and watch birds yourself. Neither precludes the other.
While I love this idea generally, the scope of the use of generative AI for artwork is too vague for my comfort.
The README.md mentions that "no art is AI-generated, though some has been retouched with AI." Superficially, that sounds mild. Some of the other closely-related projects, however, use different verbiage, or none at all.
For those of us with ethical concerns around generative AI application to artwork, much more detail may be needed before we could decide if this was appropriate for us.
Nothing at runtime uses AI either, at least in the sense of LLMs or diffusion. The collage is Pillow and numpy, and detection is BirdNET, which is a classifier, not a generative model.
However I have used LLMs for code-related work, and for writing scripts for programatically editing the images, like cutting, contrast and colour corrections.
https://doi.org/10.1016/j.ecoinf.2021.101236
It always seems kind of silly to me to throw everything at an LLM. I know they’re huge and can automatically handle a huge number of tasks but something in me finds it wasteful when we could be creating easily trainable, cheap to run bespoke models for a lot of stuff
It seems IP over Avian Carriers is finally within reach!
0. https://github.com/tphakala/birdnet-go
https://www.birds.cornell.edu/home/merlin/
I just got and setup a BTLE e-ink driver the other day and according to my calculations it will last _years_ on a single charge (2000mAh) even with multiple refreshes per day (crazy compared to the Wifi ones). Year+ lifetime completely changes the calculus for where to put these IMHO.
I'm considering designing a multiple-screen "art" piece to hang on my wall and playing around with various ideas (on both content and design).
LLMs + 3D Printer + e-ink is my new favorite hobby.
Screen: 7.5" Monochrome eInk / ePaper Display with 800x480 Pixels (https://www.seeedstudio.com/7-5-Monochrome-ePaper-Display-wi...)
BTLE board: XIAO ePaper Display Board(nRF52840) - EN05 (https://www.seeedstudio.com/XIAO-ePaper-Display-Board-nRF528...)
Battery: 3.7V 2000mAh - (https://www.amazon.com/dp/B0FR9GH966)
Picture of it all assembled (no case yet): https://cs.joshstrange.com/zJFvGGPB
The image rendered to the screen is coming from a little service I wrote that renders a book quote it pulls from BookOrbit which syncs with KOReader (annotations, progress, etc). My plan is to set up one of these one each bookshelf/series rotating quotes I highlighted from the series.
Here are 2 ESP32-based screens I have, the one on the left is a an all-in-one [1] and the one on the right I printed the case from the TRMNL DIY kit from SeeedStudio [2]: https://cs.joshstrange.com/knk4Ns2h
Right now only the BTLE one is running OpenDisplay (the other two are semi-managed by HomeAssistant ESPHome) and all the devices are in HA for controlling them, though the ESP32's all deep-sleep for saving battery so I can can't "push" stuff to them like I can the BTLE one.
[0] https://opendisplay.org/index.html#battery
[1] https://www.seeedstudio.com/reTerminal-E1001-p-6534.html
[2] https://www.seeedstudio.com/TRMNL-7-5-Inch-OG-DIY-Kit-p-6481...
The BTLE boards have just crazy higher battery time with no need for deep sleep which is why I'm moving in that direction.
~7.5" is the size I've been buying so far, the larger e-ink screens get pricey but I can't wait for a calendar-sized one to come down in price such that I can frame it and hang it on the wall. Lastly I'd avoid the color e-ink screens unless you know you need color, won't need to refresh it often, and it's not in your eye line. Color refreshes take way longer (~30s in my experience for a 5-6 color screen) and the flashing it has to do is obnoxious. Monocolor displays take <2s normally and a single flash. I got a color screen first because "why not?" and I got color, but the refresh was too annoying so now it refreshes once a day in a place that I'm not when it refreshes so I don't have to see it.
I also recommend you check out OpenDisplay for what you will run on the device: https://opendisplay.org/index.html
[0] https://www.seeedstudio.com/reTerminal-E1001-p-6534.html
Refreshes every hour
Battery lasts a few months
Esp32 and a colour eink panel, same as this one I think
https://theodore.net/store/avian-visitors/
https://github.com/Twarner491/AvianVisitors
https://theodore.net/projects/AvianVisitors/
Or maybe im missing to see is the same author? It went viral a few months ago
And just because 2 things do the same job doesn't make one a "copy" of the other.
They are awful close to still require a reference to the original project imo. They are achieving the same output (identify birds and display them in a frame display all hosted in a raspberry pi) with that twist to how to generate the art.
I understand OP is continuing the work and improving it as he sees fit, but that in my mind is a fork and should aknowledge the initial work that sparked it.
Maybe i'm wrong tho and this is completely parallel work! Waiting for OPs reply
I've seen other similar projects pop up like https://github.com/veteranbv/inky-bird-frame and https://github.com/adamoberley/HABirdDashboard/tree/HABirdDa..., each with their own spin.
I still think it would be good to acknowledge the original project that sparked the idea, especially now that you've mentioned it was one of the inspirations.
Definitely keep going with yours and evolve it further. That's how good projects grow. But for a healthy open source community, I think we should make a point of crediting the people whose work inspires us. Otherwise, over time, it can discourage people from sharing their projects publicly in the first place.
I'll leave it up to you to decide whether one should expect the demographic of this website to be especially invested in the Great Man narrative but trying to convince them that two people can independently have similar-sounding ideas involving similar-but-different technology is probably an uphill battle. Just be glad you're not composing music for a living because there's literally an entire industry built around the idea that it's impossible for two people to come up with similar melodies without knowingly or unknowingly copying eachother and that it's very important that one of them (or rather the company they had to sign a deal with in order to make any money off their music at all) 100% owns that exact melody and should be paid by anyone who wants to use it.
Next time you have an idea you should just file a patent tbh - the insistence on prior art invalidating the work seems to be quite similar but at least if you're granted a patent you can shake down "copycats" for money.
I wish E-ink displays were a bit more affordable. The 13 inch one is £229.50. Does anyone know why they are still so expensive, at least at large sizes?
Disclaimer: I am the CEO of Pimoroni :-)
The latest are pretty complex devices, very slow to refresh (hence why they are not used in your e-reader) and also very expensive. But the rendering is way much nicer because it’s real ink that is used to render the colors.
On a side note. How do people deal with all the ugly mess birds cause in such settings? I'd love to have such thing but seeing all the white stuff on glass door does not really feel fun to me overall.
You gave me few ideas for the analyzer setup too - thanks!
https://github.com/defl/hokku_epaper
I can't help wonder what this would look like for humans. Imagine, set one of these up at a party and it has the means to identify anyone on the invite list. When it hears the person it displays a photo, brief bio, and relationship status.
Amazing thing, will replicate.
A few things that might be interesting:
- The screen only redraws when the set of birds changes, and dithers the collage down to six colours for the e-ink display.
- Bigger birds sit toward the centre, scaled by real body mass.
- 800+ cutouts across 400+ species, each cut from a real public domain plate. All art is historic and human-made. Coverage is currently best for the Nordics, Britain and Germany (but other parts of the world are in the works).
- Runs fully local on a Raspberry Pi or your homelab (yes, even the classifier runs great on a RPI)
The e-ink panel is optional and it can run web-only in a container against a BirdNET-Go you already have.
Happy to answer any questions!
Right now the coverage of North America is relatively sparse, but the interest has been high and I'm hoping for some contributions.
Edit: For North America the most obvious source is Audubon's Birds of America, which is public domain and should provide a lot of coverage. This collection includes some beautiful plates.
I'm based off Spain, so I'll be happy to contribute if I find missing Iberian and Mediterranean species, or if some cool feature idea comes up :)
The README.md mentions that "no art is AI-generated, though some has been retouched with AI." Superficially, that sounds mild. Some of the other closely-related projects, however, use different verbiage, or none at all.
For those of us with ethical concerns around generative AI application to artwork, much more detail may be needed before we could decide if this was appropriate for us.
Every bird is a cutout from a real scanned 1800s plate (Gould, the von Wright brothers, Dresser...). The manifest (https://github.com/arnegiacomo/fugleramme/blob/main/assets/a...) links to each source, so you can compare them to the original scans. No bird has been prompt/diffusion generated, although I did use diffusion-based tools to remove birds from the perches (https://github.com/arnegiacomo/fugleramme/tree/main/assets/a...).
Nothing at runtime uses AI either, at least in the sense of LLMs or diffusion. The collage is Pillow and numpy, and detection is BirdNET, which is a classifier, not a generative model.
However I have used LLMs for code-related work, and for writing scripts for programatically editing the images, like cutting, contrast and colour corrections.