Summary

Telegram voice messages are received but never transcribed. The root cause is that applyMediaUnderstanding is never called during Telegram message processing.

Steps to reproduce

1. Receive a voice message via Telegram in OpenClaw
2. Voice file is saved to /Users/rain/.openclaw/media/inbound/
3. But no transcription occurs - no .txt file is created

Expected behavior

Voice messages should be automatically transcribed to text (Chinese language support needed).

Actual behavior

• Voice files are saved but never transcribed
• The applyMediaUnderstanding function is never triggered for Telegram voice messages
• This appears to be a bug in the Telegram plugin

Environment

• OpenClaw version: 2026.1.30
• OS: macOS 25.2.0
• Install method: npm/node via launchd

Root Cause Analysis

1. Missing plugin call: OpenClaw's applyMediaUnderstanding is not being called during Telegram message processing
2. No Chinese transcription support: Deepgram API (currently configured) does NOT support Chinese language - only English models available (nova-3, 2-general)
3. Workaround needed: Need OpenAI Whisper API or local Whisper for Chinese transcription

Logs

Telegram voice messages received but not transcribed
Audio format: Ogg Opus, mono, 48000Hz
Files saved to /Users/rain/.openclaw/media/inbound/ but no transcription happens

Suggested Fixes

1. Fix Telegram plugin: Ensure applyMediaUnderstanding is called when voice messages are received
2. Add Chinese support: Integrate OpenAI Whisper API (supports Chinese) as an alternative to Deepgram
3. Or support local Whisper: Allow using openai-whisper locally for transcription

Additional Context

The user has configured OpenAI API Key, but automatic transcription still doesn't work because the Telegram plugin never triggers the transcription workflow.

Labels: bug, telegram, voice, transcription