Show HN: Australian Acoustic Observatory Search

[sdenton4]

https://search.acousticobservatory.org/

The Australian Acoustic Observatory (https://acousticobservatory.org/) has 360 microphones across the continent, and over 2 million hours of audio. However, none of it is labeled: We want to make this enormous repository useful to researchers. We have found that researchers are often looking for 'hard' signals - specific call-types, birds with very little available training data, and so on. So we built an acoustic-similarity search tool, allowing researchers to provide an example of what they're looking for, which we then match against embeddings from the A2O dataset.

Here's some fun examples!

Laughing Kookaburra: <https://search.acousticobservatory.org/search/index.html?q=h...>

Pacific Koel: <https://search.acousticobservatory.org/search/index.html?q=h...>

Chiming Wedgebill: <https://search.acousticobservatory.org/search/index.html?q=h...>

How it works, in a nutshell: We use audio source separation (<https://blog.research.google/2022/01/separating-birdsong-in-...>) to pull apart the A2O data, and then run an embedding model (<https://arxiv.org/abs/2307.06292>) on each channel of the separated audio to produce a 'fingerprint' of the sound. All of this is put in a vector database with a link back to the original audio. When someone performs a search, we embed their audio, and then match against all of the embeddings in the vector database.

Right now, about 1% of the A2O data is indexed (the first minute of every recording, evenly sampled across the day). We're looking to get initial feedback and will then continue to iterate and expand coverage.

(Oh, and here's a bit of further reading: https://blog.google/intl/en-au/company-news/technology/ai-ec... )

[defrost]

The obvious and very Australian bird question is, of course, how can we search for:

* A camera shutter,

* A camera with a motorised drive,

* A car alarm,

* A chainsaw, ... etc ?

https://youtu.be/mSB71jNq-yQ?t=113

[dpeckett]

Super cool! The examples are really slick. What sort of research do you see the A2O supporting?

As an Australian living abroad I've been long fascinated by the potential for AI across the continent, you have vast areas of land where there is a tremendous lack of human labor available. It's probably a big part of why invasive species have become so difficult to control, labor intensive management and monitoring techniques just don't scale.

These days I work on industrial edge computing (increasingly focusing on ML). Super interested in the potential to get models running in the field (at scale, on cost optimized hardware). One of my favorite Aussie AI applications has to be the felixer: https://thylation.com/.

[sdenton4 → dpeckett]

There's a couple rules for the a2o in conservation. First, it provides a baseline that we can compare other data collection against - lots of projects will be shorter lived our very local in scope, and so the nearest station can provide some kind of comparison. For more wide-ranging species, we should be able to get a better sense of the geographic variation in vocalisations.

Second, and more direct for this project, there's a lot of questions in monitoring where training data is lacking for classifiers. We see the search tooling as a great way to quickly generate reference recordings to build classifiers.

[Rygian]

This is most definitely a modern-day sect of the Listening Monks.

https://wiki.lspace.org/Listening_Monks