AL·IX
A Lifeform, version IX

The wake-word loop

diagram

♪ a chime marks every mic-open and close: consent you can hear “Hey Alix” wake word fires Capture one utterance Whisper speech → text Chat turn the normal path Speak reply in the room 6-second follow-up window: no wake word needed heard nothing usable? she stays silent rather than answer static
How a spoken conversation runs: her name wakes her, one utterance is captured and transcribed, it becomes an ordinary chat turn, and she answers aloud, with a short window to keep going. · full diagram →

How a spoken conversation actually runs, on one screen.

You say her name. A listening window opens, with an audible chime, so anyone in the room hears it, one utterance is captured, an open-source speech-to-text model turns it into text, and that text runs through the exact same path as a typed message. Then she answers out loud. For about six seconds after she replies you can keep talking without saying her name again. And if she heard nothing usable, she says nothing, rather than answer static.

The full story, training a wake word on a made-up name, and the office microphone that quietly sabotaged every early attempt, is in Learning to hear.

← All entries