A LITTLE LESS SCREEN. A LITTLE MORE LISTENING.
Bring back
a sound.
Step outside for a short listening walk. Collect eight seconds of something you notice, then let a small open model suggest what to listen for next.
Your listening stop
Up to 30 seconds / 10 MB. Stay still while recording.
Reference clips from credited creators; not recorded on your walk.
What did you notice?
Your sound suggestions will appear here.
You decide what you heard. The model offers clues, not a definitive identification.
Saves your note and model suggestions locally. Audio is not saved.
YOUR NEXT LISTENING WALK
Find three layers of sound.
Take a ten-minute walk somewhere familiar. Pause safely and listen for one near sound, one far sound, and one sound you almost missed.
OUTSIDE
Pocket field journal 0
A walk, a pause, one thing you heard. Start small.
About the model and privacy
Soundwalk uses Google's open YAMNet sound-event model through MediaPipe. Recognition runs in your browser, including when you choose a file. Your recording is never sent to a server. Model scores are suggestions, not calibrated probabilities or bird-species IDs.
The first load needs internet to download the local model and runtime. After “Offline copy ready” appears, this page can reopen without a connection in browsers that support service workers. Private-site sign-in may still require internet. The upstream runtime documents performance metrics sent to Google. This page restricts network connections to its own origin to block those external requests. Notes remain in this browser; clearing site data removes them.
Built for the Hacktoberfest “Touch Grass” challenge. No claim of outdoor field testing. Open-source credits