A note to just be a bit careful passively monitoring ocean acoustics, it’s easy to fall foul of military / security forces, they don’t like anything that can fingerprint a vessel.
I worked on DAS acoustic monitoring for subsea power cables (to monitor cable health!), turns out they are basically a submarine detection system.
Last Chance to See has a fun bit about listening for dolphins in the Yangtze by taking a regular microphone and putting a condom over it. Always wondered how they sealed the end.
TIL about "plug-in power", that seems to be a thing that some sound recording devices with 3.5 mm "phono" jacks can provide.
Here [1] is a page at Klover, and here [2] is one at Shure. Not sure if there's a formal specification for this, or if it's just something that manufacturers started doing.
The most interesting bit here to me isn’t the $5 or the DIY, it’s that this is quietly the opposite of how we usually “do” sensing in 2025.
Most bioacoustics work now is: deploy a recorder, stream terabytes to the cloud, let a model find “whale = 0.93” segments, and then maybe a human listens to 3 curated clips in a slide deck. The goal is classification, not experience. The machines get the hours-long immersion that Roger Payne needed to even notice there was such a thing as a song, and humans get a CSV of detections.
A $5 hydrophone you built yourself flips that stack. You’re not going to run a transformer on it in real time, you’re going to plug it into a laptop or phone and just…listen. Long, boring, context-rich listening, exactly the thing the original discovery came from and that our current tooling optimizes away as “inefficient”.
If this stuff ever scales, I could imagine two very different futures: one is “citizen-science sensor network feeding central ML pipelines”, the other is “cheap instruments that make it normal to treat soundscapes as part of your lived environment”. The first is useful for papers. The second actually changes what people think the ocean is.
The $5 is important because it makes the second option plausible. You don’t form a relationship with a black-box $2,000 research hydrophone you’re scared to break. You do with something you built, dunked in a koi pond, and used to hear “fish kisses”. That’s the kind of interface that quietly rewires people’s intuitions about non-human worlds in a way no spectrogram ever will.
> You’re not going to run a transformer on it in real time
Why not? You can run BirdNET's model live in your browser[0]. Listen live and let the machine do the hard work of finding interesting bits[1] for later.
This is what I was going to say. My whole goal when setting up sensing projects is to eventually get it to a point that I can automate it. And I'm just a DIY dude in his house. I've been working on the detection of cars through vibrations detected by dual MPUs resonating through my house. I don't mean to imply I've had great success. I can see the pattern of an approaching car but I'm struggling to get it recognized as a car reliably and to not overcount.
But yeah, totally been doing projects like this for a long time lol not sure why OP implies you wouldn't do that. First thing I thought was "Oh man I want to put it in the lake near me and see if I can't get it detecting fish or something!"
Slight tangent, but does anyone have experience with recording hydrophones in excess of 192khz? Last I checked, most of these are specialty devices with high price tags.
Recording full-fidelity whale or dolphin sounds (amongst others) requires using a higher sample rate than is available in most consumer-grade equipment. There's a lot more information down there!
These hydrophones are a bit more expensive (~$1k per deployment) but still very accessible compared to how much it usually costs. And the goal is to bring the cost down to the ~$100 range (so $5 is very impressive!):
All the data is being saved (used for scientific research & ML training), with some of the hydrophones going back to 2017, and yes it's quite difficult to listen to and review so much audio. Better tools like the hydrophone explorer UI are much needed (been working on something similar).
One of the things that's surprised me the most is how difficult to keep hydrophones up and running. I can sympathize with both the technical and social challenges—underwater is not a friendly environment for electronics, and it can be difficult to get permission to deploy hydrophones. But it's incredibly rewarding when it works and you capture some cool sounds.
For anyone interested, all the code is open source and acoustic data is freely available:
13 comments
[ 1.9 ms ] story [ 42.3 ms ] threadI worked on DAS acoustic monitoring for subsea power cables (to monitor cable health!), turns out they are basically a submarine detection system.
Here [1] is a page at Klover, and here [2] is one at Shure. Not sure if there's a formal specification for this, or if it's just something that manufacturers started doing.
[1]: https://www.kloverproducts.com/blog/what-is-plugin-power
[2]: https://service.shure.com/s/article/difference-between-bias-...
Can we now have lot of audio records with a documentation of whale behavior to train an AI and get a whale-translator at the end?
Most bioacoustics work now is: deploy a recorder, stream terabytes to the cloud, let a model find “whale = 0.93” segments, and then maybe a human listens to 3 curated clips in a slide deck. The goal is classification, not experience. The machines get the hours-long immersion that Roger Payne needed to even notice there was such a thing as a song, and humans get a CSV of detections.
A $5 hydrophone you built yourself flips that stack. You’re not going to run a transformer on it in real time, you’re going to plug it into a laptop or phone and just…listen. Long, boring, context-rich listening, exactly the thing the original discovery came from and that our current tooling optimizes away as “inefficient”.
If this stuff ever scales, I could imagine two very different futures: one is “citizen-science sensor network feeding central ML pipelines”, the other is “cheap instruments that make it normal to treat soundscapes as part of your lived environment”. The first is useful for papers. The second actually changes what people think the ocean is.
The $5 is important because it makes the second option plausible. You don’t form a relationship with a black-box $2,000 research hydrophone you’re scared to break. You do with something you built, dunked in a koi pond, and used to hear “fish kisses”. That’s the kind of interface that quietly rewires people’s intuitions about non-human worlds in a way no spectrogram ever will.
Why not? You can run BirdNET's model live in your browser[0]. Listen live and let the machine do the hard work of finding interesting bits[1] for later.
[0] https://birdnet-team.github.io/real-time-pwa/about/
[1] Including bits that you may have missed, obvs.
But yeah, totally been doing projects like this for a long time lol not sure why OP implies you wouldn't do that. First thing I thought was "Oh man I want to put it in the lake near me and see if I can't get it detecting fish or something!"
Recording full-fidelity whale or dolphin sounds (amongst others) requires using a higher sample rate than is available in most consumer-grade equipment. There's a lot more information down there!
For example:
12 MHz (i.e. 12 000 kHz) sample rate per channel, 4 channels, 16 bit ADC:
https://www.akm.com/eu/en/products/mfp-lbp/ak8471vn/
single digit dollar unit prices:
https://octopart.com/search?q=AK8471VN¤cy=USD&specs=0
https://live.orcasound.net/
These hydrophones are a bit more expensive (~$1k per deployment) but still very accessible compared to how much it usually costs. And the goal is to bring the cost down to the ~$100 range (so $5 is very impressive!):
https://experiment.com/projects/can-low-cost-diy-hydrophones...
All the data is being saved (used for scientific research & ML training), with some of the hydrophones going back to 2017, and yes it's quite difficult to listen to and review so much audio. Better tools like the hydrophone explorer UI are much needed (been working on something similar).
One of the things that's surprised me the most is how difficult to keep hydrophones up and running. I can sympathize with both the technical and social challenges—underwater is not a friendly environment for electronics, and it can be difficult to get permission to deploy hydrophones. But it's incredibly rewarding when it works and you capture some cool sounds.
For anyone interested, all the code is open source and acoustic data is freely available:
Code: https://github.com/orcasound/
Data: https://registry.opendata.aws/orcasound/
Community: https://orcasound.zulipchat.com/