Description

Currently when using OpenClaw built-in TTS (Edge TTS) to generate audio and send via Feishu, the audio is sent as an MP3 file attachment instead of a native Feishu voice bubble.

Root Cause

Feishu API requires OGG/Opus format for native voice messages, but Edge TTS generates MP3.

Solution

Modify 5 files in openclaw-lark extension to auto-convert MP3 to OGG/Opus and send as native voice bubble.

1. messaging/outbound/media.js

• Add convertMp3ToOgg() using FFmpeg with libopus codec 64k
• In uploadAndSendMediaLark: detect .mp3, convert to .ogg, set isVoiceAudio=true
• In fetchMediaBuffer: add /tmp and backslash-tmp to defaultRoots for Windows compat
• Use sendAudioLark() for isVoiceAudio+isAudio cases

2. messaging/outbound/actions.js

• Extract asVoice from params (asVoice or audioAsVoice)
• Pass through deliverMessage and deliverMedia

3. messaging/outbound/deliver.js

• Accept and forward asVoice parameter to uploadAndSendMediaLark

4. messaging/outbound/outbound.js

• Add resolveMediaLocalRoots() with os.tmpdir() fallback
• Forward mediaLocalRoots and asVoice in sendMedia/sendPayload

5. card/reply-dispatcher.js

• Add mediaLocalRoots and asVoice in both streaming and static deliver paths

FFmpeg

FFmpeg is required for conversion. Windows install via winget install ffmpeg.

Call Chain

TTS MP3 -> reply-dispatcher -> deliver.js -> media.js (convertMp3ToOgg) -> sendAudioLark -> Feishu voice bubble

Impact

• Native Feishu voice bubbles for all TTS content
• No breaking changes (graceful fallback to file attachment)
• Tested on openclaw-lark 2026.3.31 Windows