Sound Polaroid is the handheld, physical form of the Soundscene project a one-button device that “photographs” a place by listening to it instead of looking at it.
The idea – A camera captures how a place looks; Sound Polaroid captures how it sounds and turns that into a printed souvenir. The output is deliberately not evidence or a photograph, it’s a designed interpretation of a short listening moment, ending on the printed line “A place, heard once.”
How it works:
Say where you are — the device asks for your location out loud; speech is transcribed, with the last spoken place kept in flash as a fallback.
Keep quiet / listening — a XIAO ESP32-S3 Sense with an INMP441 mic holds a rolling ~10s buffer of ambient sound, triggered by the front-panel button on GPIO 7.
Machine listening — audio is classified (YAMNet on-device, Perch/PANN server-side) into likely sound events: birds, traffic, voices, water, wind.
Interpretation made visible — detected labels are mapped through an authored visual vocabulary into a readable, inspectable prompt. Nothing is hidden behind a polished result.
Generation — the prompt (not raw audio) goes to the souvenir engine on Vercel, which returns the scene image.
Print — the device connects over BLE to a thermal printer and prints a receipt-style souvenir: heading, place and date, generated image, spectrogram, the sound names it heard, capture duration.
Games & Computer Arts 
Ethan Hay | Games & Computer Arts
