Bug

Voice messages (OGG/Opus) sent via Telegram arrive as raw audio attachments and are NOT auto-transcribed, despite having tools.media.audio properly configured.

Config

{
  "tools": {
    "media": {
      "audio": {
        "enabled": true,
        "models": [
          { "provider": "openai", "model": "gpt-4o-mini-transcribe" }
        ]
      }
    }
  }
}

Verified via openclaw config get tools.media — config is loaded correctly.

Environment

• OS: Windows 10 (10.0.19045) x64
• OpenClaw: 2026.2.19-2
• Node: v24.13.0
• Channel: Telegram (bot, working correctly for text messages)
• OpenAI API key: Valid and working (manual Whisper API transcription succeeds with the same key)

Steps to Reproduce

1. Set tools.media.audio.enabled: true with OpenAI provider model
2. Restart gateway (openclaw gateway restart)
3. Send a voice message via Telegram to the bot
4. Voice message arrives as <media:audio> with raw OGG file attachment
5. No transcription occurs — Body is not replaced with transcript

Expected Behavior

Voice message should be auto-transcribed and the transcript should replace the message body, per the docs at https://docs.openclaw.ai/nodes/audio

Workaround

Manual transcription via ffmpeg + OpenAI Whisper API works fine:

ffmpeg -y -i input.ogg output.wav
# Then POST to /v1/audio/transcriptions with whisper-1

Related

Possibly related to #7899 (Telegram voice messages not transcribed - applyMediaUnderstanding not called)