Birdsong Audio Separation

January 24, 2022 12:27 PM

Birdsong Audio Separation
We made some cool new machine learning models for separating birdsong in soundscape recordings, and demonstrated how to use the separated audio to improve downstream classification. The separation model is available on github, where we've also got lots more examples. There's also a paper.

Role: machine educator

posted by kaibutsu (4 comments total)

I imagine this could be used to count birds remotely and unattended. This is very cool.
posted by drowsy at 4:28 PM on January 27, 2022 [1 favorite]

This is remarkable - and the graphic visualisations would be very helpful in public interpretative contexts. I don't do much with birds, but very often design for their inclusion and protection.

I'm considering retweeting this as there are several NZ bird people who follow me/I follow.
posted by unearthed at 3:36 PM on January 29, 2022 [1 favorite]

Hi! Yes, we've been working with the Cal Academy of Sciences, who are using these models to understand the impact of prescribed fires on animal populations. (So far, it looks very good; prescribed fires have lower intensity, so most species stick around, and we also get for adapted species coming in.) This /should/ make it easier to justify prescribed burns, which are really some of the best protection against the giant wildfires we've seen over the last few years.

We've also made the sortation model available to help other ecologists and conservationists worth their own goals.
posted by kaibutsu at 4:05 PM on January 29, 2022 [1 favorite]

a relevant expired unresolved ask-me
posted by aniola at 9:14 AM on September 27, 2022