Universal voice: the same listening in the app, on the web and in the extension, whisper included
By Weslley Harakawa
Talking to an AI employee has to feel as natural as talking on the phone. That is why ORYKSA uses a single voice engine across every product: the app, the website chat, the dashboard and the browser extensions. The rules are the same everywhere, so the experience is the same.
How it knows you are talking
The engine listens to the microphone continuously and measures two things every fraction of a second: the level above the room noise and the periodicity of the sound in the human voice band, between 70 and 350 Hz. A voice is periodic; a key press, a door or traffic is not. The phrase opens at the first sound above the floor and closes after 700 milliseconds of silence. It is only sent if it holds at least 500 milliseconds of speech, so a cough never reaches the server, and a phrase lasts at most 15 seconds. One second of pre-roll makes sure the first syllable is never lost.
Cutting the employee
While the employee speaks, only a human voice interrupts her: 130 milliseconds of periodic sound above a fixed floor, above the echo of her own voice. When that happens she stops on the word, what she already said stays in the chat and the person's recording starts with the first second already kept.
Whisper in, whisper out
If the person whispers, the engine notices the low level and the lack of periodicity, and the reply comes back whispered, in the same voice. Spoken replies are short, up to two sentences, and the full text always stays in the chat.
Mute without going deaf
Mute stops the sound and the listening and clears every counter; resume leaves the engine ready at the first tap. The noise floor adapts to the room, with limits so it is never deaf in a cafe and never triggered by the wind.
The numbers are the ones of the app approved by the owner, and any future change goes through the app first.
