EchoFind
On-device ML for household mystery sounds. What's beeping? Where is it? EchoFind answers both.
Something in your house has been beeping for three days. You can hear it. You cannot find it. EchoFind is the mobile app for exactly that — the small, persistent, low-frequency mysteries of household sound.
What it is
EchoFind is a consumer mobile app for iOS 16+ and Android 12+ that uses on-device machine learning to classify household mystery sounds and localize their source via walk-path heatmapping. You walk through your house with the app open; EchoFind builds a heatmap of where the sound is loudest and tells you what it most likely is.
Privacy-first by design: raw audio never leaves the device. Classification happens on-device. The heatmap stays local. There is no cloud upload, no server transcript, nothing for anyone else to read.
Status
v1 in progress, iOS first. Sound classification is functional today — 48 kHz capture into YAMNet running as a Core ML model, 25 household classes, live in the app, classifying from a fresh clone because the model ships in the repo. Localization is Phase 2: a CoreMotion dead-reckoning walk-path heatmap, in progress. Android is an empty skeleton and the backend routes are stubs; both are post-v1. v1 runs on stock pretrained YAMNet, with a model fine-tuned on the weak classes landing in v1.1 — until then, low-coverage classes are badged experimental in the UI, and electrical_arc carries an explicit "not a safety device" disclaimer. Reach out if you'd like to be on the early-access list.