> generate lifelike bird sounds to enhance the samples of underrepresented species. These can then be used to train audio identification tools used in ecological monitoring
That's a pretty neat idea. "feature generation" but for rare birdsong spectrograms.
If the generative model can accurately generate sounds matching a specific underrepresented specie, it must have somewhere an embedding that does separate well and could be used for classification by clustering. No?
I have a friend who uses generative AI to classify and count wild flower species in alpine meadows using drones that autonomously patrol the same areas in many locations every day indefinitely to evaluate how climate change is impacting biodiversity in those ecosystems.
Yeah here’s a recent paper they published. They have more elaborate programs underway in the US and in Switzerland but I have no idea where to source details
Check out BirdNET [1] and BirdWeather as well. BirdNET is from Cornell University and there's a version of it for the Raspberry Pi [2]. BirdWeather [3] combines all the publicly available BirdNET instances into a nifty map view.
I took a break from software to complete a masters degree in ecology a little while ago and loved it. So much potential for software to improve that field. My research was based on using remote sensing to detect change in ecosystems, which proved to be very useful when monitoring raised peat bogs - vitally important ecosystems.
I'm back in the software world now but still spend time doing surveys and trying to think of ways I can combine both skillsets into a viable and impactful business.
>so much potential for software to improve that field
Reading Dr Fei Fei Li's "The worlds I see" really drove this home. And not just ecology, Soooo many fields would benefit from that type of interdisciplinary communication. Medicine, 100k people die a year from medical mistakes. Some would be remedied with something as simple as a sensor that detects when a patient has been still too long.
It's basically a book on the history of computer vision paired with a biography, and it's amazing to see how much early psychological, neurological work went in to the early field of computer vision. And how much cooperation needed to happen for ai to even come near healthcare. Really made me appreciate how interconnected nearly all intellectual pursuits are.
15 comments
[ 4.3 ms ] story [ 51.0 ms ] threadThat's a pretty neat idea. "feature generation" but for rare birdsong spectrograms.
That’s just awesome.
Indeed! Any links to share?
https://www.biorxiv.org/content/10.1101/2023.03.28.533305v2
[1] https://birdnet.cornell.edu/
[2] https://github.com/mcguirepr89/BirdNET-Pi
[3] https://www.birdweather.com/
So grab a spare Raspberry Pi, a GPS, a cheap USB sound card and a mic and get recording with this Pi based Acoustic Recording Unit
https://github.com/hcfman/sbts-aru
And while you are at it, install 3x or more and localize where the birds are.
I'm back in the software world now but still spend time doing surveys and trying to think of ways I can combine both skillsets into a viable and impactful business.
Reading Dr Fei Fei Li's "The worlds I see" really drove this home. And not just ecology, Soooo many fields would benefit from that type of interdisciplinary communication. Medicine, 100k people die a year from medical mistakes. Some would be remedied with something as simple as a sensor that detects when a patient has been still too long.
It's basically a book on the history of computer vision paired with a biography, and it's amazing to see how much early psychological, neurological work went in to the early field of computer vision. And how much cooperation needed to happen for ai to even come near healthcare. Really made me appreciate how interconnected nearly all intellectual pursuits are.