Subvocal's AI Powered Glasses Decode Silent Speech at 200+ WPM, Raising Privacy Questions

Y Combinator Backs Silent Speech Technology

Subvocal, a Y Combinator-backed startup founded by Georgia Tech computer science students Aarush Yadav and Arnav Kumar, has developed smart glasses that transform silent mouth movements into text without producing any sound1

1

2

2

. The AI powered glasses use low-powered radio waves projected toward the wearer's face to detect muscle movements and facial shape changes during subvocalized speech. This approach to human-computer interaction enables users to compose emails, prompts, messages, and code at the speed of speech while maintaining complete silence in their surroundings.

From Hackathon Prototype to $1.3 Million Funding

The journey began just four months before the announcement when Yadav and Kumar built their first version at Stanford's Treehacks Hackathon in February 20262

2

. Their initial prototype used EMG electrodes attached to the face, which proved impressive enough to secure early funding from San Francisco VC firm Afore Capital1

1

. The team has since raised $1.3 million and pivoted to RF sensing technology, recognizing that a commercial dictation product requiring facial electrodes would face adoption challenges. The founders took leaves of absence from Georgia Tech to pursue their startup full-time.

Technical Breakthrough in Silent Communication

The smart glasses achieve silent communication by emitting radio waves similar to Bluetooth frequency but at much lower power levels. When users mouth words, even discreetly, facial muscle movements and changes in mouth structure distort these radio waves in ways that AI models can decode into written text2

2

. The system currently operates at 200+ words per minute with what the company describes as a reasonably low error rate, though specific accuracy figures haven't been published1

1

. Getting RF sensing to work presented significant AI-driven technical challenges. The team had to build an extremely low-noise RF sensing stack and strategically position antennas so their fields overlapped maximally with the articulators—the facial muscles involved in speech production.

Minimal Calibration for Hands-Free Interaction with AI

Subvocal has pre-trained the AI models handling speech recognition to minimize setup time for new users. During a demonstration with Y Combinator partner Dave Lieb, the system required reading just 20 sentences to calibrate the headset to his specific speech patterns1

1

. This streamlined calibration process makes the technology more accessible for hands-free interaction with AI systems. The founders believe this efficiency could particularly benefit heavy AI users who type lengthy prompts frequently. "A lot of AI native people type out very long prompts to AI very often," Yadav explained. "If you have 10 agents running at the same time, you need to instruct the AI a lot."2

2

Privacy Concerns and Cognitive Implications Surface

Dr. Rafael Yuste, a Columbia University neuroscientist and director of its NeuroTechnology Center, expressed cautious optimism while raising data privacy concerns. "I am intrigued but cautious: it's a clever engineering step for human-computer interaction, but Subvocal hasn't published accuracy figures," he stated2

2

. Yuste warned that the more physiological data we generate through such devices, the greater the privacy concerns, especially if that data flows to companies or governments. He also highlighted potential cognitive consequences: "Writing and articulating words are part of how we organize and refine our thinking, and offloading all of that could weaken those processes in ways we don't yet understand." These privacy concerns extend beyond typical wearable technology as the glasses continuously monitor facial muscle movements to enable silent communication.

Preorders Open for Q1 2027 Delivery

The smart glasses are available for preorder at an early bird price of $499, with full retail pricing set at $699 and an estimated delivery in Q1 2027, specifically March 20271

1

2

2

. Early adopters pay $199 upfront, with the remaining $300 due upon shipment. The current prototype features obvious 3D-printed components, but mockups on the company's website show a sleeker final design. Key specifications like battery life, accuracy rates, and companion app details remain undisclosed. Yadav defended the technology against concerns about increased screen dependence, arguing that the glasses could actually reduce device reliance by putting computing in the background and allowing users to remain present in the moment while AI handles tasks in the cloud.