Mood
Under your mood is a state it can read: how calm or keyed-up you are, how focused, how loaded. It does not judge whether you are happy or sad.
CYMBA C1-S Concept render
Speak
without sound.
Silent speech input from a pair of earbuds and a soft boom along the jaw.
It reads the muscles you move when you mouth a word — not the words you think.
State indications · wellness, not yet validated
Mood. Stress. Sleep.
Stress
Stress and recovery trends, led by heart rate and HRV from the same electrodes and cross-checked against in-ear EEG.
Sleep
Overnight in-ear EEG gives a sleep staging indication: wake, REM, light and deep sleep.
The in-ear electrodes in both earbuds record EEG, and the same electrodes pick up the heart's electrical signal. What comes out are wellness indications, not diagnoses — and none of them has been validated on our device yet.
01
Focus and cognitive load
Changes in in-ear alpha and theta rhythms indicate how focused you are, and how loaded.
02
Stress, recovery and arousal
Led by heart-rate variability and cross-checked against in-ear alpha: a trend for stress, recovery and arousal level.
03
Sleep staging
Overnight in-ear EEG indicates wake, REM, light and deep sleep.
04
Heart rate and HRV
The heart's signal, picked up by the same electrodes, gives resting heart rate and HRV trends.
Silent speech · today and next
Today, your lips move. One day, perhaps they won't have to.
Muscle-based silent speech
You make the movements of a word without voicing it. The boom reads the muscle activity under your jaw and matches it to a short list of commands. Enrollment is required.
EEG-based silent speech is coming
Lips still, words only in your head. From non-invasive EEG, published results so far are close to chance. We are researching it, with no promised date and no promised accuracy.
Large language model · signal to text
A language model turns the signal into words.
A decoder first turns muscle or brain activity into candidate words, each with a probability. A large language model then weighs those candidates against how language really goes, and writes out the most likely text.
01 · Signal
02 · Decoder · candidates and probabilities
- play0.58
- pay0.24
- bay0.11
03 · Language model · text
Play music
EMG → text · today
Muscle signals carry the words
In published silent-speech research, adding a large language model cut open-vocabulary word errors from 28.8% to 12.2%. On C1-S the model works inside your closed set of commands and phrases.
EEG → text · research
Brain signals, checked against noise
EEG is far weaker, so the language model would be doing most of the work — and a language model will happily turn noise into a fluent sentence. Every EEG result is therefore tested twice: with the model switched off, and with pure noise fed in. If noise reads as well as the brain, nothing was read.
Below a confidence threshold, C1-S shows nothing rather than a guess.
CYMBA C1-S does not read thoughts. When you mouth a word without voicing it, the muscles under your jaw still move. Electrodes on a soft boom pick up that activity, and a decoder trained on your own articulation matches it to a short list of commands.
Closed set
Commands, digits and short phrases. Not free dictation.
Enrollment
It learns how you articulate before it works. Not out of the box.
Status
Not yet validated on people. The first study has not been run.
Signal
What the boom picks up.
- Silence. A quiet baseline under the jaw.
- You mouth a word. Short bursts of muscle activity appear on the boom channels.
- Only that short window is passed to the decoder.
Simulated4 channels shown
Generated in your browser: 1/f noise, an in-ear alpha rhythm and bursts of jaw-muscle activity. Not a recording and not human data.
Electrodes
Three places on the head. A pair of earbuds holds them.
- 01 · In the ear
A concha leaf
In each ear, three dry electrodes rest in the concha, with a reference electrode on the anti-tragus. They carry in-ear EEG for focus indication.
In-ear leaf: three electrodes and a reference.Concept render - 02 · Behind the ear
A small pod
Behind each ear, three electrodes about 7 mm apart, in a shallow zigzag, sit on the skin.
Behind-ear pod: three electrodes on the skin side.Concept render - 03 · Along the jaw
A soft boom
A boom you bend by hand attaches to either earbud and follows the jaw. Four electrodes, 12 mm apart near its tip, face the skin under the jaw. Whether this placement works is what the first study will test.
Boom: four electrodes on the skin side; bends by hand.Concept render
Modes
Two modules. One connector.
Boom
Silent speech
Attach the boom to mouth commands without making a sound.Concept render
Forehead band
Focus
Swap to the band, with four electrodes on the forehead, for focus indication. A wellness feature, not a medical one.Concept render
Dock
2 × 9 magnetic contacts
Both earbuds carry the same magnetic dock. The boom clips to either one; the band connects to both. Boom and band are never worn together.Concept render
Privacy · design intent, not yet built
Locked unless you unlock it.
01
Two deliberate steps
A double jaw clench opens a three-second window. A silently mouthed passphrase confirms it. Anything else stays locked.
02
Gated on the chip, before the radio
While locked, the radio carries no signal data at all. Not encrypted data — none.
03
Visible to the room
A light on the earbuds stays on while decoding, so people nearby can tell. Decoding closes after 60 seconds, or the moment you take an earbud out.
Concept specifications · subject to change
CYMBA C1-S
- Form
- A pair of earbuds · behind-ear pods · one module
- In the ear
- 3 dry electrodes + anti-tragus reference, in each ear
- Behind the ear
- 3 electrodes per side · ~7 mm pitch · shallow zigzag
- Boom
- 4 electrodes · 12 mm pitch · bend by hand · fits either side
- Forehead band
- 4 electrodes
- Connector
- 2 × 9 magnetic dock on each earbud, shared by boom and band
- Primary signal
- Surface EMG of silent articulation
- Secondary signals
- In-ear EEG and ECG, from the same electrodes
- Wellness indications
- Focus · cognitive load · stress, recovery, arousal · sleep staging · heart rate and HRV
- Vocabulary
- Closed set: commands, digits, short phrases
- First language
- Mandarin Chinese & English
- Enrollment
- Required for each user
- Privacy gate
- On the chip, before the radio (design intent)
- Validation
- Not yet validated on people
Family
The whole kit.
All images are AI-generated concept renders, not photographs. Electrode positions have not yet been validated on people.
CYMBA C1-S is a concept in development. It does not decode inner speech or thoughts. Silent speech input needs enrollment and works only when you really articulate, silently. State indications are wellness features, not diagnoses. Not a medical device.