Silent Voice combines lip reading and muscle signals

Several signals help choose a word

The camera reads visible mouth movement; electrical muscle sensors add another signal. A language model uses conversation context to help distinguish similar-looking words. The team describes a phone-call demonstration.

Silent Voice combines lip reading and muscle signals ↗

The demonstration has narrow limits

The repository says the demonstration used a healthy participant and reports no patient validation. Thai Tran’s account describes a roughly one-second voice delay and adaptations trained on the demonstration speaker. This does not establish reliable communication for other people.

Silent Voice combines lip reading and muscle signals ↗Silent Voice maker account, October 27 ↗

What you can do

1. Read the prototype’s validation limits and creator’s account. Open reference ↗

Where the evidence stops

No hardware assembled, call made or model executed. This is a research lead, not a clinically validated communication aid.

Sources read. Sources read November 2, 2026 UTC. No software installed, inference run or complete workflow executed.

Sources & further reading