Show HN: Subvocal – Talk to your computer without making noise
Silent-speech “smart glasses” that use low-power radio waves to read mouth and tongue movements and convert them to text have sparked both excitement and skepticism. Commenters see major potential for hands‑free input, accessibility, and dictation in noisy or public spaces, but question the glasses form factor, cloud‑only inference, privacy guarantees, long‑term RF safety, and whether accuracy will be good enough for everyday use. Many also argue the product page undersells how the technology actually works, while some doubt the ambitious shipping timeline and worry it could end up as a non‑shipping or niche device.
What Subvocal Is and How It Works
- Smart glasses that let users “talk without talking” by silently mouthing words; RF sensors in the frame capture internal mouth/tongue muscle movement and send it to cloud models for transcription.
- No cameras; earlier experiments with lip-reading, IMUs, EMG, mmWave radar, and throat mics were abandoned due to information limits, form-factor issues, or needing audible whispers.
- RF signal is low-power and only active during dictation; it interacts with different tissues as a complex dielectric, modulating phase/amplitude.
Accuracy, Latency, and Capabilities
- Current prototype reportedly achieves open‑ended silent speech decoding with ~sub‑500ms end‑to‑end latency; accuracy not yet at “product ready” and not quantified on the site, which several see as a red flag.
- Supports English only for now; other languages planned post‑launch.
- Can work with lips closed (“internal speech”) after user training, but this is harder to learn; most demos show normal mouthing with minimal visible movement.
- Speech can be transcribed in very noisy environments, and RF could also be used in the future to denoise regular audio.
Form Factor and UX Critiques
- Strong debate about glasses vs alternatives (earbuds, necklaces, caps, headbands, clip‑ons). Many dislike being forced into a particular frame style, especially those with prescriptions, big/small noses, or who don’t normally wear glasses.
- Some want detachable temples that attach to existing frames; founders say consistent antenna placement and space constraints drove the glasses choice.
- Website heavily criticized: scroll-jacked landing, vague “how it works”, lack of immediate one‑sentence explanation, and AI-generated imagery that some feel undermines trust.
Privacy, Cloud Inference, and Trust
- Decoding currently requires internet and runs in the cloud via companion apps; local inference is “planned later”.
- Multiple commenters say they will not use or buy without fully local processing, especially since this effectively replaces a keyboard (highly sensitive text).
- Founders say data is encrypted, processed in hardware isolation, and only used for training with consent; many replies remain unconvinced, citing common “we take privacy seriously” patterns.
Use Cases and Potential Impacts
- Enthusiasts highlight: faster input than typing for most people, hands‑free coding and prompting, RSI mitigation, use in offices/public transit, loud environments, and accessibility for people with speech/voice impairments.
- Others argue typing or quiet whispering is sufficient, question real‑world benefit for coders, and worry about constant AI conversations worsening social disconnection.
Safety and Health Concerns
- Several ask about RF near eyes/brain and long‑term cancer/SAR issues; founders assert safety and low power but provide no formal studies or certification references, which some find unsatisfying.
Hype, Skepticism, and “Scam” Concerns
- Many are genuinely excited, referencing sci‑fi (Ender’s Game, Doctorow, Vinge, 1984) and calling it “straight out of a novel”.
- Others doubt shipping timelines (Q1 2027), error rates, and real‑world usability, and some label it a potential scam due to AI renders, preorder model, and ambitious claims; defenders point to technical depth and prototype videos as counter‑evidence.