Bug type

Behavior bug (incorrect output/state without crash)

Summary

When an agent sends a voice note via the MEDIA: directive with an MP3 file on Telegram, the audio is delivered twice — once as a file attachment and once as a voice note. The recipient sees two separate messages containing the same audio.

Steps to reproduce

1. Configure Microsoft TTS with en-GB-ThomasNeural on OpenClaw 2026.4.11
2. Set messages.tts.auto to "off" (to isolate the MEDIA path from the auto-TTS path)
3. Generate an MP3 file using node-edge-tts and save to ~/.openclaw/workspace/.openclaw/media/voice-reply.mp3
4. Reply with MEDIA:.openclaw/media/voice-reply.mp3 as the only content
5. Observe: two separate Telegram messages are received — one as a file attachment, one as a voice note

Expected behavior

A single voice note should be delivered.

Actual behavior

Two identical audio messages are delivered to the Telegram chat:

1. First: appears as a media file attachment
2. Second: appears as a native Telegram voice note

Workaround

We worked around this by bypassing OpenClaw's MEDIA delivery entirely and calling the Telegram Bot API directly via curl -F "voice=@file" sendVoice, which delivers a single voice note correctly.

Related issues

• [Bug]: Edge TTS generates audio but sendVoice fails for model-reply TTS tags in Telegram (auto / tagged mode) #45329 — Edge TTS generates audio but sendVoice fails (network error path)
• fix(tts): allow OpenClaw temp directory paths in reply media normalizer #63511 — fix(tts): allow OpenClaw temp directory paths in reply media normalizer (fixes the /tmp path blocking, which is a separate but related bug)

Additional context

• This was reproduced on macOS (Apple Silicon), OpenClaw 2026.4.11, Telegram channel
• The duplicate delivery happens regardless of whether [[audio_as_voice]] tag is included
• The tts tool output is also blocked by the /tmp path issue (fix(tts): allow OpenClaw temp directory paths in reply media normalizer #63511), but even when using workspace paths that ARE allowed, the MEDIA directive causes duplication
• Voice: en-GB-ThomasNeural, format: audio-24khz-48kbitrate-mono-mp3

Environment

• OpenClaw: 2026.4.11 (769908e)
• OS: macOS 26.4.1 (arm64)
• Channel: Telegram
• TTS: Microsoft Edge (node-edge-tts)