quite.spaceThere is a walkie-talkie mode, and it is built on one decision: the voice itself never leaves your device. Hold the button, speak — speech is recognized on the phone, and what travels is an ordinary text message. On the other side it is spoken aloud, prefixed with who said it. Two people with phones in their pockets can talk hands-free, and what actually crossed the network is a sentence, not a recording.
Why text on the wire
- It fits everywhere. A sentence is a few hundred bytes: it rides the internet, a local network, or a LoRa frame with equal ease — a voice channel over radio at these bitrates is physically impossible, but a conversation is not.
- It stays a conversation. Everything said in radio mode lands in the space as normal messages — searchable, quotable, readable later by someone who joined after. Nothing exists only as air.
- Nothing to intercept twice. Recognition happens on the device; no audio is uploaded anywhere, no cloud speech service is involved. The model lives on your phone.
How it behaves
- Hold to talk — release, and the recognized sentence is sent as you. Recognition is fast enough to keep rhythm: well under a second for a short phrase on a mid-range phone, in both English and Russian.
- Incoming speech is spoken — “nika: we reached the camp” — through the system voice, with the sender’s name first, so a pocketed phone is a radio receiver.
- The locked screen counts. A message arriving while the screen is off is still spoken. The phone in a jacket on a trail behaves like the radio it is pretending to be.
Honest edges
Recognition quality is the gate: the on-device model holds everyday speech in RU/EN well, but it is a transcription, and it will occasionally glue words — what was said is what you see, and what you see is what traveled. Over LoRa the mode works because text is text; the radio path itself is the young part, as the radio page says plainly.