Detector spikes

Throwaway harnesses. Each one measures whether a detector is accurate and fast enough to build a drill on. Neither one teaches you anything, they only test the plumbing.

1. Voice → note name

Mode 1 input. You are told a note name, you say it, the page logs what it heard. Isolates the recognizer from your theory knowledge.

Watching for: B/C/D/E/G confusions, and engine latency measured from end of speech.

Runs Vosk locally. Fetch the model once with script/get-vosk-model.

2. Guitar → pitch class

Mode 2 input. You are asked for a note, you play it in any octave, the page logs the pitch class it detected.

Watching for: false positives from a previous note ringing out, and onset→commit latency.

Both pages need microphone access, so they need a secure context: localhost or HTTPS. A phone hitting a LAN IP over plain HTTP will be refused by the browser. They are also ES modules, so file:// will not work either.