🗣️
COMING SOON
Six Seven Voice Speedrun
The goal is still simple: say six seven as many times as possible in 20 seconds while staying clear at ridiculous speed.
The half-working prototype is parked. General ASR, wake-word spotting and an off-the-shelf word classifier all worked slowly but missed too many repetitions during continuous fast speech. The next version will use a tiny model trained specifically for this game.
⚡ Continuous speedNo phrase-ending pause and no waiting for a long transcript to settle.
🧠 Purpose-built modelIt only needs to understand six, seven and everything else.
🔒 Local inferenceThe target remains browser-local scoring without sending microphone audio to a speech-recognition service.