PROTOCOL NOTES / 10

Voice-activated transmission

Send speech, pause silence, and never cut the first syllable.

Android and iOS software implemented · Apple Wi-Fi Aware, Bluetooth LE and mixing tested on hosts and emulators only · Physical qualification pending

On by default, physically unqualified

A phone transmits only while its rider is talking. The setting “Transmit only while talking” is on by default; turning it off switches to continuous transmission at the next frame. Host tests and the speech corpus lost no speech frames or onsets; helmet, wind and natural-conversation quality remain unqualified, and no battery saving is claimed.

Words come first

The pinned WebRTC RNN VAD scores each 10 ms block after echo cancellation and noise suppression. A block transmits on voice evidence (probability at least 0.5), on a level onset 12 dB above the noise floor that is not echo, on an echo-like block within 25 dB of the rider’s learned speech level (double talk), on any missing, stale or discontinuous analysis, and during the first 2 seconds. One second of hangover follows the last such sign. A 20 ms frame pauses only if both of its blocks do.

Every unmuted frame is still Opus-encoded, with codec DTX off. The last five paused frames (100 ms) are held; when transmission resumes they are protected and sent first, oldest first, so a soft onset the detector notices late still arrives. Mute, a capture discontinuity or a new audio session discards held frames.

A quiet source still has a clock

Paused frames consume no RTP sequence number or SRTP index, so sequence numbers stay contiguous across a pause while RTP time advances. The talk-spurt marker is set on the first pre-roll frame. Signed source-status hints help receivers distinguish intentional silence from packet loss. Source resynchronization can advance progress across a partition but must not rewind replay state or permit same-SSRC key replacement.

Resume without replaying a silence backlog

Pre-roll frames arrive late on purpose: the marker frame and the next four never size the jitter target or raise it as missed playout. Receive policy uses source timing to choose FEC and bounded concealment, skips recovery across intentional gaps, and re-anchors a source that has been silent for a second. Direct ping/pong liveness continues independently, so an intentionally quiet source is not considered a dead transport. Host corpus checks are separate from helmet, wind, and simultaneous-speaker qualification.

Bridge mixes follow their sources

A bridge mix has no voice detector of its own. A mix frame is sent only while at least one included source is active, and the bridge’s own capture counts only for frames its voice encoder transmitted, pre-roll and hangover included. Mix frames have no pre-roll or hangover of their own and publish no source status. Receivers recognize their pauses from timestamp and index progression, and the pre-roll playout exemption does not apply to them.

Authority: Voice-activated transmission (DTX). Public examples are synthetic. This guide is maintained against the private implementation; it does not imply access to its source.