Bug type

Behavior bug (incorrect output/state without crash)

Summary

When messages.tts.auto: "inbound" is configured and a user sends a voice message on Matrix, the TTS reply audio is sent as a generic audio file attachment instead of a native voice message bubble (MSC3245). This is because "matrix" is not included in the VOICE_BUBBLE_CHANNELS set in src/tts/tts.ts.

Root Cause

In the TTS pipeline (src/tts/tts.ts):

const VOICE_BUBBLE_CHANNELS = new Set([
  "telegram",
  "feishu",
  "whatsapp",
]);

Matrix is missing from this set. This causes two cascading issues:

1. resolveOutputFormat("matrix") returns DEFAULT_OUTPUT with openai: "mp3" and voiceCompatible: false instead of TELEGRAM_OUTPUT with openai: "opus" and voiceCompatible: true
2. The reply dispatcher sets audioAsVoice: result.voiceCompatible === true → false, so the Matrix send pipeline never applies MSC3245 voice metadata

Even though extensions/matrix/src/matrix/send.ts fully supports audioAsVoice and correctly injects org.matrix.msc3245.voice + org.matrix.msc1767.audio, the TTS pipeline never sets this flag for Matrix.

Steps to reproduce

1. Configure messages.tts.auto: "inbound" and an OpenAI-compatible TTS provider
2. Send a voice message to the bot on Matrix (Element)
3. Bot transcribes, generates a reply, and runs TTS
4. TTS audio is sent as a generic file attachment, not a voice bubble

Expected behavior

TTS replies on Matrix should:

• Use Opus format (like Telegram/Feishu/WhatsApp)
• Set voiceCompatible: true
• Be sent as MSC3245 voice message bubbles with waveform, duration, and inline playback

Actual behavior

TTS replies on Matrix:

• Use MP3 format (DEFAULT_OUTPUT)
• Set voiceCompatible: false
• Appear as downloadable file attachments without inline playback

Suggested fix

Add "matrix" to VOICE_BUBBLE_CHANNELS:

const VOICE_BUBBLE_CHANNELS = new Set([
  "telegram",
  "feishu",
  "whatsapp",
  "matrix",  // Matrix supports MSC3245 voice messages
]);

This would make the TTS pipeline use Opus format and set voiceCompatible: true for Matrix, enabling native voice bubbles.

Related issues

• Matrix: Voice messages sent as audio files instead of native voice messages #14225 — Matrix sendMedia adapter missing audioAsVoice pass-through (different layer)
• [Bug]: Matrix message tool ignores audioAsVoice parameter #32489 — message tool audioAsVoice not wired to Matrix (different layer)
• Both related but this issue is specifically about the TTS auto-reply pipeline, not the message tool

Environment

• OpenClaw version: 2026.3.1
• Channel: Matrix (Element Web/Desktop)
• TTS provider: OpenAI-compatible (custom endpoint)

Impact

• Affected: All Matrix users receiving TTS auto-replies
• Severity: Medium — voice reply works but UX is poor (file download vs inline playback)
• Frequency: 100% for Matrix TTS replies