AudioClaw Skills Voice Intake

When to use

Use this skill when the user sends a voice message and AudioClaw should understand the content before replying.

Common triggers:

• A Feishu or chat bot receives an audio message instead of text.
• AudioClaw needs a transcript plus a clean user message payload.
• The workflow wants richer ASR features such as timestamps, sentiment, or speaker separation.
• The team wants one stable AudioClaw intake entrypoint instead of hand-written ASR requests.
• The channel stores inbound voice files as .ogg or .opus, and AudioClaw still needs one stable ASR path.

Do not use this skill for speech output. Use $audioclaw-skills-voice-reply for TTS.

Workflow

1. Save the incoming audio file locally.
2. Run scripts/openclaw_voice_intake.py with the audio path.
3. Let the script choose the best model when no model is forced:
   • sense-asr-deepthink for normal single-speaker voice understanding
   • sense-asr when a language hint is provided
   • sense-asr-pro when timestamps, sentiment, speaker diarization, or punctuation are requested
   • sense-asr-lite when hotwords are requested
4. Use the JSON manifest it returns as the AudioClaw handoff:
   • transcript.normalized_text
   • openclaw.turn_payload
   • routing.selected_model
5. If understanding.clarification_needed is true, ask the user to repeat or resend the audio.

Runtime model

Official HTTP ASR API:

• Endpoint: https://api.senseaudio.cn/v1/audio/transcriptions
• Content type: multipart/form-data
• File size limit: \x3C=10MB
• Practical local input suffixes accepted by this skill: .wav, .mp3, .ogg, .opus, .flac, .aac, .m4a, .mp4

Supported response goals:

• plain transcript
• richer raw response passthrough
• AudioClaw-ready turn payload

The skill keeps two layers separate:

• ASR output from AudioClaw ASR
• AudioClaw packaging and clarification heuristics

API key lookup

This skill now treats SENSEAUDIO_API_KEY as the default API key source again.

Runtime rules:

• If the host app injects SENSEAUDIO_API_KEY as an AudioClaw login token such as v2.public..., the shared bootstrap will replace it with the real sk-... value from ~/.audioclaw/workspace/state/senseaudio_credentials.json before ASR starts.
• --api-key-env still works, but the default runtime path is SENSEAUDIO_API_KEY.

Commands

Basic voice intake:

python3 scripts/openclaw_voice_intake.py \
  --input /path/to/user_audio.mp3

Voice intake with richer AudioClaw structure:

python3 scripts/openclaw_voice_intake.py \
  --input /path/to/meeting_clip.m4a \
  --enable-punctuation \
  --timestamp-granularity segment \
  --enable-sentiment \
  --out-json /tmp/openclaw_voice_turn.json

Force a specific model:

python3 scripts/openclaw_voice_intake.py \
  --input /path/to/user_audio.mp3 \
  --model sense-asr-deepthink

AudioClaw integration pattern

Recommended handoff:

1. Channel adapter stores the inbound audio.
2. AudioClaw calls scripts/openclaw_voice_intake.py.
3. AudioClaw reads:
   • openclaw.turn_payload.role
   • openclaw.turn_payload.content
   • openclaw.turn_payload.metadata
4. The normal dialogue pipeline continues as if the user typed the recognized text.

Operational rules:

• Keep the original audio path in metadata for debugging.
• Pass language only when you are confident; otherwise let ASR auto-detect.
• If you request timestamps, sentiment, or diarization, let the script choose sense-asr-pro.
• If transcript is empty, do not hallucinate a user intent. Ask for clarification.

Resources

• scripts/senseaudio_asr_client.py
  • Multipart HTTP client for AudioClaw ASR
  • Handles model routing validation and JSON or text responses
• scripts/openclaw_voice_intake.py
  • Main runtime for AudioClaw
  • Builds transcript, normalized user text, and turn payload
• references/openclaw_voice_intake.md
  • Official ASR docs summary, model support notes, and AudioClaw payload examples