A practical use for AI: generative metadata for music!
Developer: Source Audio
Standard tagging = genre, instruments, mood, has vocals, tempo, style, keywords
cost: $0.20 per song
Weird it doesn’t include musical key…
How long until we see this available for all sounds?
Similarly it would be useful to have AI analyse a photo archive and auto assign metadata and description… It would sure make finding photos on your phone a lot easier. But I wonder if that would then mean the analysed media then becomes a part of the training data? If so, its a strange business model…
I’ve been using ‘Any Vision’ for tagging in Lightroom: https://johnrellis.com/lightroom/anyvision.htm
It’s far from meticulous hand curation, and I don’t use some features – identifying Landmarks (only picked up really ‘famous’ ones i.e. super touristy / international ones), Faces (don’t take enough portraits with my ‘big camera’ that’s all on my phone), Safety (smut filter) – but overall it’s pretty impressive at object and scene type identification and when it stuffs up it’s usually for obvious reasons when you look at the image e.g. it’s abstract, esoteric or ambiguous.
The AI processing is farmed out to Google’s ‘Cloud Vision’ so initial setup is a bit of a mission. And you have to turn your credit card number over to Google but, as you manually select the batches of images to send for tagging and there’s a generous number of free lookups initially plus and an ongoing monthly free quota, I’ve never paid more than a few dollars when I wanted to run a large number through in one-shot.
And while some might balk at the Google side of things the developer’s own licensing approach can’t be faulted:
“Buy a license at a price you think is fair:
The license includes unlimited upgrades. Make sure you’re satisfied with the free trial before buying.”
Fascinating, thanks!
Friendly greetings Tim! Do you use Google Photos or Apple’s built-in one? Both of those do tagging under-the-hood, so you can search by people/places/things in the photos. YES it does help to train their systems (per their fine print).
AI image generators like Midjourney let you upload an image to “describe” it, too.
Also https://edenphotos.io/
As part of my work, I looked into a lot of AI music tagging not long ago. You may also be interested in these:
– https://www.disco.ac/
– https://cyanite.ai/
– https://www.bridge.audio/
– https://www.aimsapi.com/
Splice is definitely doing something with this kind of tech too, and their “find similar sounds”. Sononym comes to mind as well, “When a sound is added to a Sononym library, it will automatically be classified & categorized according to our machine-learning model”: https://www.sononym.net/docs/manual/categories/
Enjoy exploring! 🤘🤖🎵
Thanks! I don’t use either, but I figure sooner or later my phone will tag photos as I take them.