Hardware
The device is the second act.
Software ships first because it gets BabelGO into real conversations fastest. But the limits we keep hitting are physical, and physical limits need hardware.
Why
Four things a phone will never do well.
A phone cannot hear one person in a hall
Handset microphones are tuned for a face 20 cm away in a quiet room. A trade show floor is neither. Six microphones and real beamforming solve what software cannot.
Holding a screen between two faces kills the conversation
The moment a phone comes up, the exchange stops being between two people and starts being about the device. The best translator is one nobody looks at.
A working day is not a demo
Continuous audio capture destroys a phone battery in under two hours. A dedicated device budgets for fourteen and charges in its case.
The network is not always there
A dedicated NPU runs core language pairs entirely on-device — the same reason the pack works in a basement, a plane, or a rural clinic.
Target specifications
What we are building to.
- Microphone array
- 6-mic beamformingAdaptive noise floor tracking
- Battery life
- 14 hoursContinuous capture, 3 charges in case
- Weight
- 18 gPer side, silicone-tipped
- Inference
- On-device NPUCore pairs run without a network
- Connectivity
- BLE 5.4 + LE AudioMultipoint, two host devices
- Rating
- IP54Dust and splash resistant
These are engineering targets for the first device, published because we would rather be held to them than quietly revise them later. Where we miss, we will say so on the roadmap.