AVIF: The Avian Image Format

A rough, pixelated outline of a bird is shown in white in the top of the image. A red replica of this image is shown in a spectrogram in the lower half of the image. A smaller picture-in-picture display in the bottom right of the image shows a man sitting in a studio.

Humans have long admired the sound of birdsong, but to fully appreciate how technically amazing it is, you need an ultrasonic microphone. [Benn Jordan] recently created a video about using these microphones to analyze a collection of bird calls, even training a starling to repeat an image encoded in sound, and has some recommendations for amateurs wanting to get started in computational ornithology.

In the first part of the video, [Benn] set up automated ultrasonic recorders at home, made recordings in Florida and rural Georgia, and visited a starling named “The Mouth,” famous for his ability to mimic human sounds. As a demonstration of his abilities, [Benn] drew a simple bird shape in a spectrogram, converted it into sound, and played it for The Mouth several times. Initially, it didn’t seem that the starling would repeat it, but while he was analyzing his recordings later, [Benn] found the characteristic bird shape. The Mouth had been able to repeat it almost pitch-perfectly. It was in this analysis that the ultrasonic microphones showed their worth, since they were able to slow down the birds’ complex vocalizations enough to detect their complex structures without losing audio quality.

In the rest of the video, [Benn] shares his recommendations for recording and analyzing bird calls. He has some advice for good high-speed audio interfaces, including warnings about those that are overpriced or advertise unrealistic specifications. You’ll also need a microphone with good ultrasonic performance, and he gives a few options for this, including making your own. For analysis software, he particularly recommended Birdnet-Pi, an AI program for identifying birds by their calls, as well as Cornell Lab’s free libraries of ornithology data.

One particularly emphatic recommendation was the open-source AudioMoth ultrasonic microphone and recorder, a project we’ve seen before. He also recommended a sonic camera which serves as a more field-ready version of his own acoustic imaging device.

29 thoughts on “AVIF: The Avian Image Format

      1. I wonder if you could get it to reproduce in a scannable state… Opens up a lot of possibilities. As a backup, could just teach them to sing a dial-up modem handshake type thing with the data embedded

  1. I’m now wondering if a bird can store a zx spectrum program or perhaps even a sstv image.
    You might not be able to run doom from a bird call but the single line of BASIC version of Tetris might be possible. How many birds would it take to load doom, that is the question.

  2. Writers take note, a MacGuffin that listens to a specific bird song to activate would be pretty cool.

    Like taking a specific species from a specific place and its song opens a locked door.

        1. A probe appears in orbit over the planet and transmits a signal, which it will not cease transmitting until it receives a response that only the whales are able to provide, at which point it goes away. How is that not EXACTLY an example of challenge/response key-exchange authentication?

  3. From the video, it seems he’s identified some of the limitations of the bird’s physical ability to reproduce sounds. If you actually had an interest in storing information “in a bird” then you should incorporate these limitations into the data format/encoder to maximize the data fidelity.

    Just a thought.

    1. RIAA correction but for birds not phonographs. That would be a pretty fun project, and probably similar to the RIAA circuit in that it would bias towards higher frequency as that’s what the recording/playback medium hardware does best.

        1. You spent longer writing this inane comment that just searching for the answer.

          Who could possibly think that the recording industry association of America could have been involved in establishing industry standards related to recording?

        2. In much the same way that the FCC or FDA (and countless others) are basically a big universal access port for various lobbies, they also have a very minor function of actually setting some standards and regulations. But mostly they take bribes and facilitate cartel-like behavior

  4. So, is the bird able to whistle/produce 4 different frequencies at a time (like in the png) and even 5 (in the spetrogram one can see an additional one) ?

    Or is it some illusion due to some horizontal stretching of the spectrogram ? In reality the bird would sing one note, then another one…then repeat. And the stretching makes it look like 4 (5) continuous tones what is really a succession of 5 individual notes quickly executed.

      1. Great scrabble word by the way. Think of a recorder versus a clarinet; our larynx needs to have physical contact between vibrating surfaces, whereas with birds it’s the walls of the tube and the air column itself which vibrate. They do not have vocal chords (or the reed in this analogy). This is also why a creature smaller than your hand with lungs the size of a couple peanuts can produce a sound that nearly blows out your eardrums if it’s sitting on your shoulder

        How do they keep from ruining their own ears? There’s actually a lot of avian adaptations concerning that; birds have extremely durable ears and regenerate their hearing instead of losing it permanently over time like we do.

      1. The Fremen of Arrakis, and presumably assassins used distrans animals extensively as a form of covert communication. Typically, birds were the most common form of distrans employed in the days of the old Imperium. Bats were especially useful, as they could travel under the cover of night.

  5. There was an amazing program out there, SoundID (not the one by Sonarworks), that could recognize birdsong or any other sound using local hardware and software. Overlapping sounds, recognize at 300x realtime, signals 40 dB below the noise floor…Truly an amazing system. They used it to identify the unknown song of a bird by literally recognizing every other call

    I wanted to use it to recognize defects in an end of line tester but my organization couldn’t get off dead center, then COVID hit and the main person behind it has retired. If anyone knows of a similar system I am highly interested. I have the beta version but our plant will only install software with support.

    https://web.archive.org/web/20221013115259/https://soundid.net/
    Neal Boucher and Michihiro Jinnai were the names associated with the software.

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.