'%3e%3c!--%20Cymbal%20disc,%20top%20view%20--%3e%3ccircle%20cx='32'%20cy='30'%20r='28'%20fill='url(%23sunset)'%20/%3e%3c!--%20Lathe%20grooves:%20two%20wide%20bands%20--%3e%3ccircle%20cx='32'%20cy='30'%20r='21.5'%20fill='none'%20stroke='%23FFE3C9'%20stroke-opacity='0.55'%20stroke-width='5'%20/%3e%3ccircle%20cx='32'%20cy='30'%20r='12.5'%20fill='none'%20stroke='%23FFE3C9'%20stroke-opacity='0.55'%20stroke-width='5'%20/%3e%3c!--%20Bell%20--%3e%3ccircle%20cx='32'%20cy='30'%20r='7'%20fill='%23F5B98C'%20/%3e%3c!--%20Center%20hole%20--%3e%3ccircle%20cx='32'%20cy='30'%20r='1.9'%20fill='%23191919'%20/%3e%3c/g%3e%3c/svg%3e)
About SeaWav
A free library of openly licensed sounds, laid out on a map where similar sounds sit next to each other — so finding the right one is a matter of exploring, not scrolling.
What SeaWav does
SeaWav collects drum one-shots and other samples that are already publicly available under open licenses, listens to each of them with a machine-learning model, and places them on a two-dimensional map. Sounds that are alike end up close together: kicks cluster with kicks, bright hats drift away from dark ones, and the odd, in-between sounds live in the gaps.
How to use the map
Everything happens on the one-shot page. Each dot is a sample, colored by its class; the closer two dots are, the more alike they sound. Click any dot to hear it and hover to see its name. Right-click to open its detail card with the waveform, key, bitrate, artist and license. Press Shift to save a sound to your list, then download the whole list as a single zip. Scroll to zoom at the cursor, drag to pan, middle-click to reset the view, and use the legend or the search box to highlight a class, a key, or a name. The arrow keys jump to the nearest sample in that direction and Space replays the selection. Everything plays straight from the browser — nothing to install.
How the map is made
The layout isn't drawn by hand — it comes out of the audio itself. First the raw waveform is turned into a spectrogram, a picture of which frequencies are present at each moment. A neural network trained on large amounts of audio then reads that picture and summarizes it as an embedding: a vector of a few hundred numbers that captures timbre, attack, brightness and decay, so that two sounds which sound alike get vectors that are close together. Because a few hundred dimensions can't be shown on a screen, a neighborhood-preserving projection (the UMAP / t-SNE family of algorithms) flattens the vectors to two, keeping each sample next to the ones it was already close to. Those two numbers are the dot's position on the map. Classes and musical keys come from the source kits' metadata and are checked with automatic classification and pitch detection; the color of a dot is its class.
Where the sounds come from
Every sample comes from an openly licensed source (CC0, CC BY or CC BY-SA) or from a community submission that we review before publishing. Each sound keeps its original license, shown on its card, and can be reported if something is wrong with it.
What's next
One-shots are live today. Fills and full beats are being collected and mapped the same way and will land in their own sections soon.