Mouth the words. The frame turns them into text. No voice, no camera on your face, nothing stuck to your skin.
Who needs this
Speech without sound, for people who can't or shouldn't use their voice.
The sensor question
Every silent-speech device lives or dies on its sensor.
Tiny speakers on the frame emit chirps above hearing (18 to 39 kHz); mics read the echoes off the face. No skin contact, no camera, and the sensor moves with the glasses. Cornell's EchoSpeech showed 4.5% word error on 31 commands.
How it works
Inaudible chirps, read back as a picture of your mouth moving.
Mouth a word:Silent: the echo sits still.
Traction so far
4,033
silent-speech samples recorded on the sEMG prototype
1
closed-vocabulary decoder that works, single speaker
~$100
parts for the acoustic rig on an off-the-shelf frame
Illustration on the previous slide is simulated. Real echo data comes from the rig.
Plan
Prove the sensor before building the product.
01 Replicate
02 31 commands
03 Open corpus
04 Kill test
Rebuild published eyewear echo sensing on our own frame. Pass mark: a moving hand shows up; ten silent repeats of one word look alike.
Somach Systems, Inc.
One founder, San Francisco. A research bet with a written kill rule.