AVCF: Analog Voice Capture Format

March 28, 2026 · View on GitHub

Date: 2026-03-28 Status: Approved

Problem

The .dvcf format captures pre-vocoder codec frames for digital P25/DMR calls, making them self-contained for ASR dataset building. But trunk-recorder also records analog (conventional FM) calls, and those recordings have no equivalent self-contained format — they're just .wav files with no embedded metadata.

Users building ASR training datasets for radio transmissions need both digital and analog calls. Analog calls should be just as easy to batch-upload as digital calls: one file per call, audio + metadata included.

Design Decisions

Decision	Choice	Rationale
Unified or sibling format?	Sibling (`.avcf`)	Digital and analog go through completely different processing pipelines (IMBE-ASR vs standard ASR). Separate formats, separate plugins, separate upload paths.
Audio payload	Opaque blob with MIME type	Container is waveform-agnostic (WAV, FLAC, M4A, Vorbis). Content-type string identifies the format. No transcoding.
Plugin architecture	Separate plugin (`mqtt_avcf`)	Different data acquisition pattern — no incremental streaming, everything at call_end. Simpler than mqtt_dvcf. No risk of breaking the digital path.
MQTT payload	Full .avcf as base64	Matches mqtt_dvcf pattern (`audio_avcf_base64` + `metadata`). Consistent with mqtt_status plugin's `audio_wav_base64` convention.

AVCF File Format

Built on SSSP v2 framing (same as .dvcf). Reuses existing message types unchanged and adds one new message type.

File Structure

1. CALL_START      (msg_type=0x02, binary — identical to .dvcf)
2. AUDIO_DATA      (msg_type=0x06, content-type + audio blob — NEW)
3. CALL_METADATA   (msg_type=0x05, JSON — identical to .dvcf)
4. CALL_END        (msg_type=0x03, binary — identical to .dvcf)

Receivers must dispatch on msg_type, not position.

Shared Message Types

CALL_START (0x02), CALL_METADATA (0x05), and CALL_END (0x03) are defined in the DVCF spec and used identically here. See DVCF_SPEC.md §3.3, §3.6, and §3.4.

New Message Type: AUDIO_DATA (msg_type = 0x06)

Carries an opaque audio file with a MIME content-type identifier.

[8 bytes: SSSP v2 header]
  magic[0]    = 0x53 ('S')
  magic[1]    = 0x59 ('Y')
  version     = 0x02
  msg_type    = 0x06
  payload_len = N (uint32, little-endian)

[1 byte: content_type_len (M)]
[M bytes: content_type (UTF-8 MIME string, e.g. "audio/wav")]
[N-1-M bytes: raw audio file bytes]

Content-type values:

Extension	MIME Type	Notes
`.wav`	`audio/wav`	Default trunk-recorder output
`.m4a`	`audio/mp4`	AAC compressed
`.flac`	`audio/flac`	Lossless compressed
`.ogg`	`audio/ogg`	Vorbis compressed

The audio bytes are the raw file contents — no transcoding, no reframing. A consumer extracts them and writes directly to a file with the appropriate extension.

Rules:

content_type_len must be > 0 (content type is required).
Receivers that don't recognize the content type should skip the message via payload_len.
The audio blob is the complete file (including any format-specific headers like WAV RIFF headers).

File Extension and MIME Type

Extension: .avcf (Analog Voice Capture Format)
MIME type: application/x-avcf (unregistered)

File Naming Convention

Same convention as .dvcf — sidecar alongside the audio file:

9173-1774473273.232_852687500.0-call_39.wav   (audio)
9173-1774473273.232_852687500.0-call_39.avcf  (analog voice capture)

Callback	Purpose
`parse_config()`	Read config (same schema as mqtt_dvcf)
`start()`	Create MQTT connection if enabled
`stop()`	Disconnect MQTT, cleanup
`call_end(Call_Data_t)`	All work: read audio, build .avcf, write/publish

call_start and voice_codec_data are not used.

call_end Flow

Filter: Skip if not analog (check call_info.audio_type or analog flag). Skip if call_info.filename is empty.
Read audio file: Read the completed audio file from call_info.filename into memory.
Detect content type: Map file extension to MIME type (".wav" → "audio/wav", etc.).
Build .avcf in memory:
- Emit CALL_START from call_info fields
- Emit AUDIO_DATA (content-type + audio bytes)
- Emit CALL_METADATA (JSON from Call_Data_t, identical to mqtt_dvcf)
- Emit CALL_END from call_info fields
Write file (if write_enabled): Write .avcf sidecar (same base name, .avcf extension).
Publish MQTT (if mqtt_enabled): Base64-encode and publish to {topic}/avcf.

Configuration

{
  "name": "mqtt_avcf",
  "library": "libmqtt_avcf",
  "write_enabled": true,
  "mqtt_enabled": false,
  "analog_only": true,
  "broker": "tcp://localhost:1883",
  "topic": "trunk-recorder",
  "clientid": "avcf-handler",
  "username": "",
  "password": "",
  "qos": 0
}

Option	Default	Description
`write_enabled`	`true`	Write `.avcf` sidecar files to disk
`mqtt_enabled`	`false`	Publish to MQTT broker
`analog_only`	`true`	Only process analog calls (skip digital)
`broker`	`tcp://localhost:1883`	MQTT broker URL
`topic`	`trunk-recorder`	MQTT topic prefix (publishes to `{topic}/avcf`)
`clientid`	`avcf-handler`	MQTT client ID
`username`	`""`	MQTT username (optional)
`password`	`""`	MQTT password (optional)
`qos`	`0`	MQTT QoS level

MQTT Payload

Published to {topic}/avcf:

{
  "audio_avcf_base64": "<base64-encoded .avcf file>",
  "metadata": {
    "talkgroup": 9170,
    "talkgroup_tag": "Fire Dispatch",
    "talkgroup_alpha_tag": "BUTCO FD",
    "talkgroup_group": "Fire",
    "freq": 855737500,
    "start_time": 1711234567,
    "stop_time": 1711234590,
    "call_length": 23,
    "signal": -42.5,
    "noise": -110.2,
    "freq_error": 12,
    "spike_count": 0,
    "emergency": false,
    "priority": 3,
    "phase2_tdma": false,
    "tdma_slot": 0,
    "analog": true,
    "audio_type": "analog",
    "short_name": "butco",
    "filename": "9170-1711234567_855737500.wav",
    "srcList": [
      {"src": 1234567, "time": 1711234567, "pos": 0.0, "emergency": 0, "signal_system": "", "tag": ""}
    ]
  }
}

Cross-Project Changes

tr-plugin-dvcf (this repo)

| 0x06         | `audio_data`   | Opaque audio blob (see AVCF spec) |

Update reserved range to 0x07–0xFF.

New repo: tr-plugin-avcf

mqtt_avcf.cc — plugin source
CMakeLists.txt — same dependencies as mqtt_dvcf
AVCF_SPEC.md — this spec
README.md — build instructions, config, usage

Downstream: tr-engine

Needs a new route: audio_avcf_base64 on {topic}/avcf → standard ASR provider (Whisper, etc.), not IMBE-ASR.

No changes: IMBE-ASR

IMBE-ASR only consumes .dvcf files. Analog calls are not relevant.