That's essentially what he has done. Except he did the modulation/demodulation with audio software (and, technically, stored a monochrome bitmap, not a PNG).
Dial-up modems encode data in audio-frequency. Later modems used phase-shift keying¹, but the very early ones used frequency-shift keying², which is essentially encoding data in a frequency graph - i.e., drawing a line in a spectrum analyzer.
Drawing a bird in a spectrum analyzer is packing much more data than that; it's like playing several of those streams at once.
The bird has shown itself to be capable of remembering and reproducing multiplexed frequency-keyed streams.
>With enough birds I imagine you can store quite a bit of data. Takes saving to the cloud to another level.
Literally a point made in the video³ at 18:34.
____________
¹ https://en.wikipedia.org/wiki/Phase-shift_keying
It's a decent Fermi estimate¹ :)
We also don't know how many songs we can get the bird to memorize for us.
_________
Thinking about it in reverse, how much data would it take to encode 1 second of birdsong in the most efficient audio codec I can imagine. If M4A or MP3 with the bitrate slammed way down isn't a fair comparison, then some birdsong-specific ML autoencoder... Probably 500 bytes? Would still be enough for a Twitter tweet.
A tweet within a tweet!
I Can Run Doom On A Bird
Literally made me laugh out loud.