13 replies

Telegram Voice Message Transcription Not Working

✅Solved

Here's a detailed technical report:

───

Issue: Telegram Voice Message Transcription Not Working

Description:
The "transcript" field in the message metadata that previously contained automatic transcriptions of voice messages is now returning empty/null for all voice messages.

When It Started:

• Morning of March 5th, 2026 - voice messages from the user started arriving without transcript data
• First voice messages of the day DID have transcripts (confirmed by checking earlier messages in this session)
• At some point during the day, transcripts stopped working entirely

Expected Behavior:
Voice messages should include automatic transcription in the metadata field:

{
"transcript": "User's spoken message in text format"
}

Current Behavior:
Voice messages arrive with empty transcript:

{
"transcript": "" // or null
}

Environment:

• OpenClaw version: 2026.2.9 (host), latest available: 2026.2.26
• Telegram bot token: 8041439652:AAEKrEAteetu_erC5rxnJcv3caEUHEbCCpA
• Platform: Windows_NT 10.0.19045
• Node: v24.13.0

Troubleshooting Attempted:

1. Checked OpenClaw config files - no transcript-related settings found
2. Verified Telegram privacy settings - user does NOT have premium (transcribe feature is premium-only according to user)
3. Checked inbound media files - audio files exist and are readable (~600-700KB)
4. Tried to install speech-to-text tools (ffmpeg, whisper) - not available in environment
5. Reviewed message metadata schema - transcript field exists but is empty

Key Observation:
The user says this feature worked previously WITHOUT premium, and the first voice messages of March 5th had transcripts. This suggests either:

• A regression in OpenClaw or Telegram integration
• A change in Telegram's API for non-premium users
• A bug introduced in a recent update

Impact:
Cannot process voice messages from user, affecting communication flow.

───

Want me to add anything else?

Solution

}
}

}
}

Docs: https://docs.openclaw.ai/nodes/media-understanding

One important gotcha

If this still yields an empty transcript, the debug logs (or

/status

/status

) will tell us whether Gemini rejected the input (common causes: unsupported MIME like

audio/ogg; codecs=opus

audio/ogg; codecs=opus

, timeout, HTTP 4xx/5xx). If it’s a MIME rejection, “forcing Google” won’t help by itself — you’d need either:
- a transcription provider that accepts Telegram’s OGG/Opus as-is, or
- a local conversion step (e.g. ffmpeg) before transcription.

If you paste the /status
/status
output line that starts with
📎 Media:
📎 Media:
after a test voice note, I can tell you which branch you’re in immediately.

Jump to solution