openclaw/docs/nodes/audio.md

summary, read_when, title

summary	read_when	title
How inbound audio/voice notes are downloaded, transcribed, and injected into replies	Changing audio transcription or media handling	Audio and voice notes

What it does

When audio understanding is enabled (or auto-detected), OpenClaw:

1. Locates the first audio attachment (local path or URL) and downloads it if needed.
2. Enforces maxBytes before sending to each model entry.
3. Runs the first eligible model entry in order (provider or CLI); if an entry fails or skips (size/timeout), the next entry is tried.
4. On success, replaces Body with an [Audio] block and sets {{Transcript}}.

When transcription succeeds, CommandBody/RawBody are also set to the transcript so slash commands still work. With --verbose, logs show when transcription runs and when it replaces the body.

Auto-detection (default)

If you have not configured models and tools.media.audio.enabled is not false, OpenClaw auto-detects in this order and stops at the first working option:

1. Active reply model, when its provider supports audio understanding.
2. Configured provider auth — any models.providers.* entry with auth available for a provider that supports audio transcription. This is checked before local CLIs, so a configured API key always wins over a local binary on PATH. Provider priority when multiple are configured: Groq, OpenAI, xAI, Deepgram, Google, SenseAudio, ElevenLabs, Mistral.
3. Local CLIs (only if no provider auth resolved). OpenClaw builds an ordered fallback list:
   • whisper-cli, before CPU defaults only when an earlier model invocation in the current process observed Metal or CUDA
   • sherpa-onnx-offline on its default CPU provider (requires SHERPA_ONNX_MODEL_DIR with tokens.txt, encoder.onnx, decoder.onnx, and joiner.onnx)
   • whisper-cli when Metal/CUDA is only build-capable or the selected backend is otherwise unobserved
   • parakeet-mlx on Apple Silicon (MLX-capable; device use remains unobserved)
   • whisper (Python CLI; downloads models automatically)

Install/link provenance is capability evidence, not execution evidence. It never moves a candidate ahead of CPU sherpa by itself. OpenClaw does not load a model during setup or status checks just to probe a backend. Auto-detected whisper.cpp keeps its normal model-run logs enabled so OpenClaw can record the upstream using … backend line. Explicit CLI entries keep their configured output flags.

Gemini CLI auto-detect for media understanding was replaced by a sandboxed Antigravity CLI (agy) fallback for image/video; audio does not use a CLI fallback beyond the local binaries above.

To disable auto-detection, set tools.media.audio.enabled: false. To customize, add capability-tagged entries to tools.media.models.

Binary detection is best-effort across macOS/Linux/Windows. Make sure the CLI is on `PATH` (`~` is expanded), or set an explicit CLI model with a full command path.

Inspect the local selection without transcribing audio:

openclaw capability audio providers
openclaw doctor --lint --only core/doctor/local-audio-acceleration --severity-min info

The provider inventory reports the local fallback winner separately from global provider selection, plus capable, requested, and observed backend fields. After transcription runs, /status reports the requested or observed backend in the media line. Explicit audio-capable tools.media.models CLI entries still bypass auto-selection; use their backend-specific flags such as sherpa --provider=cuda or whisper.cpp --no-gpu/--device.

Config examples

Provider + CLI fallback (OpenAI + Whisper CLI)

{
  tools: {
    media: {
      models: [
        { provider: "openai", model: "gpt-4o-transcribe", capabilities: ["audio"] },
        {
          type: "cli",
          command: "whisper",
          args: ["--model", "base", "{{MediaPath}}"],
          timeoutSeconds: 45,
          capabilities: ["audio"],
        },
      ],
      audio: { enabled: true, preferredModel: "openai/gpt-4o-transcribe" },
    },
  },
}

Provider-only (Deepgram)

{
  tools: {
    media: {
      models: [{ provider: "deepgram", model: "nova-3", capabilities: ["audio"] }],
      audio: { enabled: true },
    },
  },
}

Provider-only (Mistral Voxtral)

{
  tools: {
    media: {
      models: [{ provider: "mistral", model: "voxtral-mini-latest", capabilities: ["audio"] }],
      audio: { enabled: true },
    },
  },
}

Provider-only (SenseAudio)

{
  tools: {
    media: {
      models: [
        {
          provider: "senseaudio",
          model: "senseaudio-asr-pro-1.5-260319",
          capabilities: ["audio"],
        },
      ],
      audio: { enabled: true },
    },
  },
}

Echo transcript to chat (opt-in)

{
  tools: {
    media: {
      audio: {
        enabled: true,
        echoTranscript: true,
        echoFormat: '📝 "{transcript}"',
      },
    },
  },
}

Notes and limits

• Provider auth follows the standard model auth order (auth profiles, env vars, models.providers.*.apiKey).
• Groq setup details: Groq.
• Deepgram picks up DEEPGRAM_API_KEY when provider: "deepgram" is used. Setup details: Deepgram.
• Mistral setup details: Mistral.
• SenseAudio picks up SENSEAUDIO_API_KEY when provider: "senseaudio" is used. Setup details: SenseAudio.
• Audio providers can use defaults under tools.media.audio or override baseUrl, headers, providerOptions, and limits on their tools.media.models[] entry.
• The built-in audio size cap is 20MB. An entry-level maxBytes override can change it; oversize audio is skipped for that model and the next entry is tried.
• Audio files below 1024 bytes are skipped before provider/CLI transcription.
• Default maxChars for audio is unset (full transcript). Set tools.media.audio.maxChars or per-entry maxChars to trim output.
• OpenAI auto-detect default is gpt-4o-transcribe; set model: "gpt-4o-mini-transcribe" for a cheaper/faster option.
• Transcript is available to templates as {{Transcript}}.
• tools.media.audio.echoTranscript is off by default; echoFormat accepts a {transcript} placeholder.
• CLI stdout is capped at 5MB; keep CLI output concise.
• CLI args should use {{MediaPath}} for the local audio file path. Run openclaw doctor --fix to migrate deprecated {input} placeholders from older audio.transcription.command configs (retired key: audio.transcription, replaced by tools.media.models).
• tools.media.concurrency bounds media tasks; it is not a GPU scheduler.

Resident local STT

Auto-detected local STT remains process-per-request. OpenClaw does not currently manage a resident whisper.cpp server because the standard Homebrew whisper-cpp package disables that server, while the upstream example has no configured bounded admission queue. A plugin-owned resident lifecycle needs a maintained packaged worker with health/startup, model residency, bounded queueing, cancellation/timeout, loopback-only no-auth operation, and no cloud fallback before it can be enabled safely.

Proxy environment support

Provider-based audio transcription honors standard outbound proxy env vars, matching undici's EnvHttpProxyAgent semantics:

• HTTPS_PROXY / https_proxy
• HTTP_PROXY / http_proxy
• ALL_PROXY / all_proxy

Lowercase variables take precedence over uppercase; NO_PROXY/no_proxy entries (hostnames, *.suffix, or host:port) bypass the proxy. If no proxy env vars are set, direct egress is used. If proxy setup fails (malformed URL), OpenClaw logs a warning and falls back to direct fetch.

Mention detection in groups

On channels that support audio preflight, OpenClaw transcribes audio before checking for mentions when requireMention: true is set for a group chat. This lets a captionless voice note pass the mention gate when its transcript contains a configured mention pattern. Channel-specific docs describe transports that require a typed mention instead.

How it works:

1. If a voice message has no text body and the group requires mentions, OpenClaw performs a preflight transcription of the first audio attachment.
2. The transcript is checked for mention patterns (for example @BotName, emoji triggers).
3. If a mention is found, the message proceeds through the full reply pipeline.

Fallback behavior: if preflight transcription fails (timeout, API error, etc.), the message falls back to text-only mention detection so mixed messages (text + audio) are never dropped.

Opt-out per Telegram group/topic:

• Set channels.telegram.groups.<chatId>.disableAudioPreflight: true to skip preflight transcript mention checks for that group.
• Set channels.telegram.groups.<chatId>.topics.<threadId>.disableAudioPreflight to override per-topic (true to skip, false to force-enable).
• Default is false (preflight enabled when mention-gated conditions match).

Example: a user sends a voice note saying "Hey @Claude, what's the weather?" in a Telegram group with requireMention: true. The voice note is transcribed, the mention is detected, and the agent replies.

Gotchas

• Scope rules use first-match-wins; chatType is normalized to direct, group, or channel.
• Ensure your CLI exits 0 and prints plain text; JSON output needs to be massaged via jq -r .text.
• Known file-output modes are authoritative: an empty or missing inferred transcript file produces no transcript instead of falling back to CLI progress output.
• For parakeet-mlx, use --output-format txt (or all) with --output-dir and the default {filename} output template. The upstream PARAKEET_OUTPUT_FORMAT and PARAKEET_OUTPUT_TEMPLATE environment variables are also honored. OpenClaw reads <output-dir>/<media-basename>.txt; the default srt format, other formats, and custom output templates continue to use stdout.
• Keep timeouts reasonable (timeoutSeconds, default 60s) to avoid blocking the reply queue.
• Preflight transcription only processes the first audio attachment for mention detection. Additional audio attachments are processed during the main media-understanding phase.

Related

• Media understanding
• Talk mode
• Voice wake