IN DEVELOPMENT
WORDLESS

Getting through without making a sound

Talking to an AI still means talking out loud. On a train, mid-meeting, beside someone asleep — the moments you most need a hand are the ones where you can't speak. That's the wall we're taking down.

HOW IT WORKS

It looks, compares, then speaks

Five steps from an expression to a single sentence — what Eclipse does in between, and what it deliberately doesn't.

More paths end early than reach the end. A faint signal, an expression that merely passed across your face, or a word already spoken not long ago — each means it lets the moment go. Staying quiet is the default, not speaking up.

THREE LAYERS

Approached in three layers

No single technique does all of it. So it's split — each layer does a different job, and each is at a different stage.

  1. 01
    Working now

    Your face reports your state

    Worn down, low, tense, or brightened — Eclipse notices and speaks first. It reads the hour and the situation to choose how to say it.

  2. 02
    Measured: 81.7%

    Your mouth signals the start

    Move your lips and it knows you're about to speak. No button, no wake word.

  3. 03
    In research

    A whisper carries the content

    Speak below what the person next to you can hear, and only that is understood. When this layer lands, soundless conversation works.

WHAT WE MEASURED

We tell you what doesn't work, first

We measured before building. One three-minute experiment ruled out weeks of the wrong direction.

01

Reading words from lips alone failed

Six words mouthed silently, then told apart: 18.3% accurate — no better than guessing (17%).

02

And we found out why

The method compresses a face into 52 numbers, and the fine differences that separate sounds are lost in that compression. The approach was wrong, not the goal.

03

One thing survived

Detecting “about to speak” works at 81.7%. Words belong to another layer; this one can be the switch.

Showing only what works looks better today and disappoints later. Saying plainly what works and what doesn't lasts longer.

WHERE IT MATTERS

The moments this would change

Not situations where you cannot speak — situations where speaking is awkward. They come up more often than you'd think.

On a train

Quietly, among people

Check the day and leave a note where saying it out loud would be awkward.

Mid-meeting

Without breaking the thread

Listen and quietly have things organised on the side. No need to step out.

Late at night

Beside someone asleep

No light, no sound — the thought that just surfaced gets kept anyway.

While driving

Your face speaks first

If you look worn, Eclipse says something first. Hands and eyes stay where they are.

⚠️The driving case is **not a safety device.** It can miss drowsiness, and missing it is dangerous — so it does not take on that responsibility. Think of it as a word from the passenger seat. Anything more needs dedicated hardware and a different standard of proof.

Still in the works
  • Whisper recognition
  • Korean lip reading
  • On your phone
  • Driving detection

Meet your own ADMM today

ADMM is getting ready. Leave your email and you’ll be the first to know the moment it opens.

We’ll only email you about the launch. Unsubscribe anytime.