OCEAN
Listen
How it works

From a creek to your screen.

An underwater microphone listens, software notices the interesting sounds, and this site shows you what it heard, honestly labelled. Here's each step, and what we can and can't know from it.

1 · The path a sound takes

Three steps, and one way in.

1

Listen

An underwater microphone hears the creek. Until ours is in the water, a real research recording of wild dolphins stands in.

2

Notice

Software turns the sound into pictures, cuts a short clip you can play, and flags sounds that look like dolphin clicks or whistles.

3

Show

Everything arrives here a second or two later, labelled with what it is: a simulation, a software detection, or the scientists' own label.

Research recordingwild dolphins · 250,000 samples a secondHydrophonein the creekRecorderon the docksoundpowerTimestampedaudiothe one way inreplayed · running todaylive · nextPipelinepictures of sound,detectors, clipsThis sitelabelledSimulationeventssaved upif offlinethe same code for a recording or the creek

The research recording and the real creek come in through the same door. Everything after it, the pictures, the detectors and this site, is the same code either way. That's why we could build all of it before anything went in the water. Solid lines are running today; dashed lines are coming next.

2 · What a recorder can hear

Faster recorders hear higher.

A recorder measures the water pressure thousands of times a second. It can only hold sounds up to half that number. Fish and people live low, where any recorder works. Dolphin clicks live far above human hearing, which is why the recorder speed matters so much. Tap a speed.

Measuring 192,000 times a second, a recorder can hold sounds up to 96 kHz. Of the creek sounds below, it catches 11, cuts off 1 and misses 1.

10 Hz100 Hz1 kHz10 kHz100 kHzLOW PITCH →→ HIGH PITCHFISHBlack drumcaughtRed drumcaughtOyster toadfishcaughtSpotted seatroutcaughtSilver perchcaughtAtlantic croakercaughtSHRIMPSnapping shrimpcaughtDOLPHINSWhistlescaughtClicks (average)cut offClicks (peak)missedOTHER SOUNDSOutboard boatcaughtShip or ferrycaughtRain (approx.)caughtPEOPLEHuman hearinglimit 96 kHz

Pitch runs left to right, each gridline ten times higher than the last. Bands are approximate, from South Carolina studies collected in our species research. The two click rows are two sources that disagree about dolphin clicks; our own 250,000-a-second research recording can settle it.

3 · Detected isn't confirmed

A machine noticing isn't a sighting.

Software can only get a sound as far as “Detected”. A person decides the rest. And “Unknown” is a kept answer, not a failure: if nobody can say what made a sound, we say so.

A soundKnownscientists' own labelsDetectedsoftware noticed itRevieweda person listenedConfirmedit is what it saysRejectedit was a boat, or a clankUnknownnobody can say yet · keptnoticeslistensscores itwith a confidence, never a fact
Known
The scientists who made the recording labelled this sound. We use their labels to check our own software, not to decide anything.
Detected
Our software noticed something, with a confidence. No person has listened yet.
Reviewed
A person listened and left a verdict.
Confirmed
Reviewed, and agreed to be what it says. Only a person can get a sound this far.
4 · Inside the hydrophone

The microphone we're building.

It's built from a $65 kit designed by a retired Navy submariner: a ceramic tube, a tiny circuit board, and resin poured around them. We'll test it in a tub, then soak it for a week and weigh it to check that no water got in, before it goes anywhere near the creek.

sound in the waterCable to the recordera little power goes down, the signal comes upA thumbnail-sized circuit boardmakes the ceramic's tiny signal strong enough to travelA ceramic tubesound squeezes it and it makes a voltage; it rings near 43 kHzCasting resinabout as dense as seawater, so sound passes through

A drawing, not to scale.

5 · For the curious

The technical details.

How this page gets its numbers

  1. 01
    Water
    replaced by an authentic research recording
  2. 02
    Hydrophone
    replaced by an authentic research recording
  3. 03
    Ocean Node
    recorder and computer on the dock: 192,000 samples a second, 384,000 planned
  4. 04
    Analysis
    spectrograms, previews, detectors run on the node
  5. 05
    Event
    known, detected, reviewed or confirmed
  6. 06
    Web
    this page, a second or two behind
Research recording
Bottlenose dolphins, NW Mediterranean, 250,000 samples a second, CC BY 4.0 (Chaparro et al. 2025)
Pictures of sound
Spectrograms made on the node from the original rate: FFT 2048, Hann window, 0–125 kHz
What you hear
A 48,000-samples-a-second preview of each second, the part people can hear
Detectors
Two transparent detectors, one for click trains and one for whistles, scored against the scientists' labels and compared with PAMGuard, the standard research tool
If the Internet drops
The node keeps recording and saves everything up, then sends it when the connection returns
The creek hardware
Planned: a hydrophone we cast ourselves and a dock-mounted recorder at 192,000 samples a second, with 384,000 as the upgrade

See it running on the Listen page, or take the dive.