Birds sounds visualized

June 18, 2017 6:09 AM

Google has had thousands of bird sounds visualized using AI. Background.
posted by Foci for Analysis (23 comments total)

Wonderful project (really!), but some caveats are in order about the way this is reported. Machine learning and Artifical intelligence work need to be really careful about tempering the kind of speculation or inference the casual consumer might draw.

The goal is classification of species (a human goal, not an avian goal). This depends on the use of species as platonic entities that must ignore variation within a species.

The language used of the machine learning system (hereafter ML) is deceptive. Loose talk of the algorithm "listening to the sounds" projects sentience onto a blind dumb number crunching algorithm.

A specific parametric representation is chosen, and this is then faded into the background to suggest that it is suited to identifying a "finger print". To most people this will remind them of real fingerprints and of shazam and suggest that essence or identity has been captured, when in fact all it means is that the representation supports the chosen exercise (human, catagorisation). The parametric representation chosen, whether Fourier spectral based, or wavelet, will have been designed to make some distinctions and ignore others. A lot of such work is based on insights from human speech research (hence the ubiquitous spectrograms), and is heavily biased towards making evident precisely those distinctions that have proven useful in the analysis of speech. This may or may not fit bird ecologies, and bird ecologies will differ, one from the other.

One severe limit of the scheme chosen is the use of one second time intervals. This makes sense (or at least, seems to work well enough) for the chosen goal, but it is a choice that does no justice to avian lives. Neither bird audition nor the ecological context in which any given sound is made enter into the process, so it must be cautious about any inference beyond the stated goal of supporting classification by humans into the ideal type of species.

I don't want to criticise the work or the intention. It is genuinely good work, and an important project, but it would be helpful if readers were encouraged not to think of algorithms as sentient, intelligent, or a nature as a book that merely needs to be read.
posted by stonepharisee at 6:31 AM on June 18, 2017 [20 favorites]

honestly i just like to push buttons to hear the birdies go chirp chirp
posted by Foci for Analysis at 6:42 AM on June 18, 2017 [16 favorites]

Just FYI, can't use it on a tablet. Off to turn on the desktop....
posted by cooker girl at 6:51 AM on June 18, 2017

Where's the one for "hey, its four a.m.! It's four a.m.! Its four a.m.! Can you believe it? It's four a.m.!"
posted by Molesome at 7:16 AM on June 18, 2017 [7 favorites]

Click and hold somewhere off center for birdgeddon.
posted by Wolfdog at 7:27 AM on June 18, 2017 [3 favorites]

It's cool and all, but, in my experience, birds often have longer calls than the short chirps given here.
posted by Thorzdad at 7:31 AM on June 18, 2017

Would make great ringtone alerts.
posted by atomicmedia at 7:39 AM on June 18, 2017 [2 favorites]

Strangely enough, the cry of the red tailed hawk made my parakeets quiet down. :)
posted by Splunge at 7:47 AM on June 18, 2017 [1 favorite]

Where's the one for "hey, its four a.m.! It's four a.m.! Its four a.m.! Can you believe it? It's four a.m.!"

In my area, that's the Eastern Whip-poor-will. You can tell it's a Whip-poor-will because of the way it is.
posted by matrixclown at 7:57 AM on June 18, 2017

It is sort of neat, though I admit to having trouble understanding it. Most of us birdwatchers -and I'm sure the Cornell lab knows this of course - identify birds not just by tone and loudness (man, can that little wren project itself!) but also by rhythm, length, time of day and of course where the call is coming from.

Not to mention that individuals vary within a species and calls vary for an individual. We have different names for different blackbirds in our area based on their song. There's the Gianni Schicchi bird because he repeats a phrase from that opera. There's the Pete Reed bird because he hangs out in the yard of that Olympian who lives near us, the Firestation bird more because of where he perches. And so on...

There's a little chaffinch who sings all day near our window. The call is very distinctive and one people recognize easily. He has two main calls though and switches between them during the day. I was just listening to him yesterday and counting how many repetitions of one call before switching...4,8-2,5-2,4 is part of what I wrote down. That is, he favors the second call much more...

Anyways, I've gone off on a tangent. All I meant to say is that these projects are neat but people should also really get out and pay attention to birdsong.
posted by vacapinta at 8:05 AM on June 18, 2017 [7 favorites]

I wish there were a scroll function on this. There also seem to be a number of repeats.
posted by EmpressCallipygos at 8:42 AM on June 18, 2017

If you just pretend that AI stands for "Available Information," tech company marketing makes way more sense.
posted by oulipian at 9:59 AM on June 18, 2017 [8 favorites]

agree with stonepharisee. This seems typical of these kind of things. Where the representation schema is masquerading as content.
posted by mary8nne at 12:09 PM on June 18, 2017

If you zoom out so all the samples fit in the screen, then wiggle the mouse about while holding the button down, someone nearby will surely ask you to stop doing that.
posted by scruss at 12:32 PM on June 18, 2017 [3 favorites]

Cool concept, but, wow, is that interface garbage.
posted by Confess, Fletch at 12:45 PM on June 18, 2017

I really didn't mean to down play what they have done. Just to highlight that it is the construction of a representation that serves specific purposes, viz. classification. A fine goal, and useful too. But it does not reveal nature as she is written, which one might otherwise surmise.
posted by stonepharisee at 12:47 PM on June 18, 2017

>If you zoom out so all the samples fit in the screen, then wiggle the mouse about while holding the button down, someone nearby will surely ask you to stop doing that.

That is so fun. You can play it like a drum. Sampling representations for creative purposes is as old as music itself. This should show up as part of a DJ set or an public jam interface.
posted by stonepharisee at 12:50 PM on June 18, 2017

If you use the interface to find Le Comte's Thrasher, and scroll around it, every time you hit it it pops out like a cork.
posted by stonepharisee at 12:51 PM on June 18, 2017

My favorite is the Lesser Nighthawk. What a crazy sound for a bird to make!
posted by Monkeymoo at 12:52 PM on June 18, 2017

Can't find the mockingbird that sounds like a car alarm. All night long.
posted by Splunge at 1:05 PM on June 18, 2017

It's better than a mockingbird that sounds like a Lionel Richie song, Splunge.
posted by scruss at 3:09 PM on June 18, 2017

I'm going to write a Hangman game and call it AI. Why not? Everyone else can call random bits of code 'AI' these days it would seem.
posted by GallonOfAlan at 6:28 AM on June 19, 2017

Would make great ringtone alerts.
I've had birdsongs as my ringtones for years. They cut through everything but aren't bothersome. Highly recommended.
posted by bink at 3:54 PM on June 23, 2017