I just did a quick check and there are quite a large number of species with 100+ sound observations: https://www.inaturalist.org/observations?place_id=97394&sounds&subview=map&view=species. Are there plans to do AI identification of them like we do for photos?
Not currently, but it’s something we’ve discussed. Definitely something we’d like to do.
I’m curious whether it would involve the development of a new tool specific to iNaturalist, or – not an exclusive “or”, I suppose – incorporation of some of the existing tools that can identify audio. people seem to rely very much on Merlin for bird sound identification, for example, though I’m not sure what the possibilities would be for welding that to sound ID for frogs, or crickets, or dolphins, or what have you.
Merlin uses basically the same technology for both sound and image recognition because it converts the sounds into sonograms.
https://www.mammalwatching.com/resources/mammal-vocalisations-audio-library/ is keeping track of the observations in iNat for mammals that include sound recordings.
Having AI sound recognition for non-birds would be tremendous. There are some resources for mammals cropping up, but it’s still scarce. The problem with sounds is that it’s impossible to look quickly at a lot of them and build some intuition, because the sound is extended in time, so having a computer do that for you would be cool.
I agree. Using Merlin, I’ve often tried to record a calling chipmunk thinking it was a bird.
Seeing a sonogram above the recording would be a very interesting and helpful feature here on iNat, don’t you think?
I usually add sonograms manually after making screenshots of a software… But it would be also nice to see sonograms on other observations…
Speaking of sonograms - as far as cicadas (Cicadoidea) are concerned - sound recordings and sonograms are as important as pictures and sometimes even more crucial. Sound is the key to the ID ;)
What do you think?
All we need is a spectrogram image, like Birdnet or Merlin produces, and a way to categorize them so they don’t muddy the waters of images used for visual ID. Should be a modest development scope.
said the person who doesn’t have to do the work…
This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.