every key has its own click · learn a few keys · watch the model identify them by sound alone · educational · runs locally
mechanical and membrane keyboards produce unique frequency signatures for each key. the sound radiates through the case and the desk, varying in timbre by position — keys in the centre sound different to corner keys. this demo captures each click, extracts a 32-bin spectral fingerprint, and matches it against a library you build yourself by pressing labelled keys. it uses nearest-centroid classification — no server, no ml library.
in “learn” mode: select a key letter, then press that physical key ~5 times · then switch to “classify” and press any learned key — watch the prediction