advertisement
advertisement

How a Cornell scientist created ‘Shazam for birds’

Grant Van Horn, a researcher at the Cornell Ornithology Lab, used artificial intelligence to identify bird calls.

How a Cornell scientist created ‘Shazam for birds’
[Illustration: German Gonzalez]

I haven’t found a cooler community to work with than nature enthusiasts. For whatever reason, it just attracts really cool people. It’s motivating to interact with them, learn how they do their visual or auditory classification, or their research if they’re card-carrying academics.

advertisement
advertisement

From December 2020 to June 2021, I was with my parents in California. It was kind of a stereotypical story, working out of my parents’ spare bedroom (creating the sound ID function on the Merlin Bird ID app, a Shazam for birdsong, which was released in July). I had the final model ready by May. That’s a fast turnaround time for something like this. That drained me mentally. The good press we got was definitely a lift, to show that it was worth it.

My best brainstorming time is not on a computer. If I’m deep in a problem—typically there’s some bug in the code and I really want to solve it—if I go on a bike ride, I’ll come back and quickly solve it, and then move on to productive work.

I went on a bike ride this weekend, and I was trying to brainstorm: If you wanted [the sound ID feature on Merlin] to be running continuously, how much power could you be drawing off your iPhone, or your watch? And is it even realistic? I would have loved to have had a summary, along my bike trip, of where different birds were singing or calling, and what species they were. I think that’s one of the beauties of audio that I didn’t appreciate until I started digging into it, how easy it is to put a microphone somewhere and just collect this ambient recording.

advertisement
advertisement

The one thing I like to emphasize is how crucial humans are to data collection—the annotation, the evaluation. Ultimately, the apps are for human consumption and have to be engaging. The machines are definitely cool, and the problems are fun to think about, but it’s the people who make all of this possible.

Time he wakes up

5:55 a.m.

First thing he does in the morning

“Brush my teeth. Then, it’s the same breakfast nearly every day: two eggs, sunny-side up, on a piece of toast or bagel or tortilla, whatever we have. My most productive hours are between 6 and 10 a.m. That’s when I want to be writing code.”

advertisement

Productivity tools

“I’m a little old school in [that] I really just want a basic code editor, and I want everything else out of my way. Because I’m not part of a big team, I’m responsible for my own deadlines. I don’t need a task manager to remind me what these things are. So give me an editor and a keyboard, and let me chip away at these things.”

Best habit, and worst

“[My worst is that] I can easily go for long hours basically bashing my head against the wall, trying to solve something. That dovetails with one of my best, which is this ability to focus. I can work for four hours straight on some decent math problems to get something to work.”

One thing he’s done this year to try to improve himself

“My dad was laid off at the end of last year. That was part of the reason I went home, because I knew I could spend more time with him. Spending time with family has been probably one of the things that’s become deliberately more important in my life. I now feel comfortable talking to Cornell and being like: Guys, I’m going to work remotely for a few months. Maybe a silver lining of the pandemic is these conversations with employers are a lot easier to have, because it’s proven how working remotely, at least in my case, it’s been shown not to be a detriment to productivity.”

advertisement

Last thing he does at night

“Turn on an audiobook or podcast. Right now, I’m listening to Andy Weir’s Project Hail Mary.”

Time he goes to bed

9:30 or 10 p.m.

Read more about the secrets of this year’s most productive people.

advertisement
advertisement
advertisement