Silent Speech Enables AI Communication Through Silently Mouthed Words

Every big computing shift comes with its own interface, and interfaces.inc is betting the AI era's interface won't be a keyboard, a mouse, or even your voice out loud. It'll be your mouth moving with no sound at all. The company just introduced Silent Speech, a system that watches how your face moves as you silently mouth words and converts that into text and commands for any AI, running on the Mac or phone you already own.

The pitch is straightforward: voice is already the fastest way millions of people talk to AI each week, but it falls apart the moment you're in an office, a meeting, or on a train. Silent Speech is meant to fill that gap, letting people mouth words instead of speaking them, then have that turned into commands, texts, or AI queries without a sound leaving their mouth.

What's notable is the technical claim behind it. Instead of relying on audio, the model works purely off pixels, tracking facial movement to guess what's being said. According to interfaces.inc, top voice models that do get full audio land around a 3% word error rate. Their pixel-only system is already under 10%, with their best result hitting 4%, which is a tight gap for a method that ignores sound entirely.

The company frames this as step one of a bigger thesis: that intelligence itself is no longer the hard problem, and the real challenge is building an interface that lets AI blend into daily life rather than sit behind a screen. Silent Speech is meant to be usable with AirPods in, mouthing a question while walking, texting without touching a screen, or getting an answer without ever making a sound.

For now, access is limited. interfaces.inc says it will invite people gradually as its research preview grows, and it's also opening the door to other companies that want to license the model, API, or SDK for voice-first products that need to work in places where actually speaking isn't an option.