openclaw - 💡(How to fix) Fix [Feature]: Feishu TTS Audio Replies Send File Path Instead of Playable Audio [1 comments, 2 participants]

Summary

When the Feishu channel receives a voice message and OpenClaw's built-in TTS auto mechanism generates an audio reply, the file path (e.g., C:\Users\...\media\outbound\voice-xxx.mp3) is sent as plain text instead of using Feishu's msg_type: "audio" API to deliver a directly-playable audio message.

Problem to solve

Current pain: Users who send voice messages expect to receive voice replies. Instead, they receive a text message containing the local filesystem path to the generated MP3 file, which is not playable and requires manual work to locate and open the file.

Why current behavior is insufficient:

• The TTS file is successfully generated at ~/.openclaw/media/outbound/voice-*.mp3
• Feishu's plugin has a working sendAudioLark() function that can send audio messages via msg_type: "audio"
• However, the built-in TTS auto delivery path does not invoke this function — it treats the audio file path as plain text and sends it via sendTextLark()
• Users receive: C:\Users\name\.openclaw\media\outbound\voice-xxx.mp3 (unplayable text)
• Users expect: 📎 audio bubble in Feishu that plays directly

Proposed solution

Option A (Recommended — Built-in): Extend OpenClaw's TTS auto mechanism to detect Feishu as the target channel and route audio files through Feishu's audio API:

1. TTS auto trigger fires after agent reply is generated
2. Detect current channel is Feishu
3. TTS generates MP3 file to media/outbound/
4. Instead of appending file path to text reply, invoke:
   • uploadFileLark() → get file_key
   • sendAudioLark() → send msg_type: "audio" message
5. User receives playable audio bubble in Feishu

Option B (Alternative): Expose a feishu_send_audio Agent Tool so the Agent can explicitly call the existing send_audio.js logic after generating TTS output, as a skill/tool step rather than automatic behavior.

Alternatives considered

No response

Impact

• Affected users/systems/channels: All users running OpenClaw with Feishu channel + TTS enabled (messages.tts.auto: "inbound")
• Severity: Annoying — blocks the intended voice reply experience but does not crash
• Frequency: Every voice message interaction where TTS reply is expected
• Consequence: Users must manually open the audio file from the filesystem path, defeating the purpose of voice reply; degrades user experience on voice-forward workflows
• Workaround exists: Manual node send_audio.js invocation works but requires the Agent to explicitly call it, which is not part of the natural TTS auto flow

Evidence/examples

Environment

• OpenClaw version: 2026.4.23
• Feishu plugin version: 2026.4.1
• OS: Windows
• TTS Provider: Microsoft (zh-CN-XiaoxiaoNeural)
• Channel: Feishu (WebSocket mode)

Related Files

• Workaround script: ~/.openclaw/workspace/send_audio.js
• Feishu audio function: extensions/openclaw-lark/src/messaging/outbound/media.js → sendAudioLark()

Additional information

No response