openclaw - 💡(How to fix) Fix TTS mode:final causes duplicate text message delivery on chat channels [1 comments, 2 participants]

Description

When messages.tts.mode is set to "final", every assistant reply is delivered twice as two separate messages on chat channels (confirmed on webchat and Feishu). The first message contains text only, and the second (arriving 6-20 seconds later) contains the same text plus the TTS audio attachment. This makes it appear to the user that the bot is repeating itself.

Environment

• OpenClaw: v2026.4.23 (a979721)
• Model: zai/glm-5.1
• TTS Provider: local-tts (Qwen3-TTS 1.7B 8-bit via mlx-audio)
• Channel: webchat + Feishu
• Config: messages.tts.enabled: true, messages.tts.mode: "final"

Evidence

Session JSONL shows two distinct assistant messages with different IDs and timestamps, but identical text content:

[503] id=0657b6c1 ts=2026-04-27T13:27:08.695Z role=assistant
     已更新。去飞书发条消息测试吧...

[504] id=d4ba053e ts=2026-04-27T13:27:14.690Z role=assistant
     已更新。去飞书发条消息测试吧...

The time delta (~6s, ~21s in other cases) corresponds exactly to the TTS synthesis duration.

Expected Behavior

In mode: "final", the text message should be delivered once, with the TTS audio appended to the same message (or sent as a separate voice-only message) rather than re-delivering the full text content a second time.

Suggested Fix

When TTS completes in mode: "final":

• Option A: Replace the original text-only message with a text+audio message (edit-in-place)
• Option B: Send only the audio attachment as a second message, without duplicating the text
• Option C: Delay text delivery until TTS is ready, then send both together as a single message

Workaround

Disable auto TTS (messages.tts.enabled: false) and use the tts tool manually when voice output is needed.