Voice store-and-forward

Voice is modeled as codec frames carried in fragment clusters (same messageId as file transfer).

VOICE_META (0x41)

Field	Size	Description
codecType	2	See registry below
sampleRateHz	2	e.g. 8000
durationMs	4	Total reconstructed audio
frameSizeBytes	2	Per codec frame
titleLen	1	Optional UTF-8 title
titleUtf8	variable

Codec registry (examples)

These codecType values are inside the VOICE_META body (typically uint16, big-endian in dumps). They are not the mesh payloadType byte.

codecType	Codec	Typical bitrate
0x0001	PCM s16le mono	N/A
0x0101	CODEC2 1200	1200 bps
0x0102	CODEC2 700	700 bps
0x0201	Opus	8–64 kbps

0x0101 is the canonical ultra-narrowband mesh voice profile for long-range links.

VOICE_FRAME (0x42)

Field	Size
frameSeq	2
frameBytes	remainder

frameSeq MUST be monotonic for the message. Missing frames trigger fragment recovery.

Store-and-forward behavior

Repeaters and gateways MAY cache complete voice messages for replay to late joiners (policy). Clients SHOULD NOT cache long audio to protect flash.

Bandwidth note

Voice competes with flood traffic. Implementations SHOULD prefer unicast / tighter TTL toward known recipients when possible and lower voice bitrate on busy default community (channelHash == 0x0000) channels.

Example flow

1. Originator sends VOICE_META + N × VOICE_FRAME fragments with messageId = M.
2. Receiver plays incrementally if codecType supports streaming; else buffers until complete.
3. If frame 7 missing, receiver issues FRAG_NACK listing index 7.