Bug Description

The tts tool creates MP3 files but they are empty (0 bytes). The audio file is not delivered to Telegram.

Environment

• OpenClaw version: 2026.3.8 (3caab92)
• OS: Linux (CachyOS, kernel 6.19.6-2-cachyos)
• Node: v25.6.1
• Channel: Telegram

Steps to Reproduce

1. In a Telegram DM with OpenClaw, agent uses tts tool
2. Observe that a voice message is sent but contains no audio (or fails silently)

Evidence

TTS tool creates empty files:

/tmp/openclaw/tts-CJ48ow/:
-rw-r--r-- 1 t7kai t7kai   0  3月11日 21:45 voice-1773236725758.mp3

Direct node-edge-tts CLI works correctly:

$ npx node-edge-tts -t "你好测试" -f /tmp/test-tts.mp3 -v zh-CN-XiaoxiaoNeural
Converted to /tmp/test-tts.mp3

$ ls -la /tmp/test-tts.mp3
-rw-r--r-- 1 t7kai t7kai 9648  3月11日 21:46 /tmp/test-tts.mp3

Expected Behavior

TTS tool should generate valid MP3 audio and deliver it to the Telegram chat.

Actual Behavior

TTS tool creates a 0-byte MP3 file. The file is empty and no audio is played.

Additional Context

• Chinese voice tested: zh-CN-XiaoxiaoNeural
• node-edge-tts npm package version: 1.2.10
• The underlying node-edge-tts library works fine when called directly via CLI
• Issue appears to be in how OpenClaw invokes the TTS internally