Description

Discord voice messages generated from a 24kHz TTS source play back at roughly 0.5x speed (noticeably slow). The same MP3 file sent as a regular attachment plays at normal speed.

Environment

• OpenClaw latest (npm)
• TTS provider: OpenAI-compatible (mlx-audio server)
• TTS model: mlx-community/Qwen3-TTS-12Hz-0.6B-CustomVoice-8bit
• Source audio: MP3, 24000 Hz, mono, 128 kbps
• Channel: Discord

Steps to Reproduce

1. Configure TTS with an OpenAI-compatible server that outputs 24kHz MP3 (e.g. mlx-audio)
2. Use the tts tool to generate a voice message
3. Listen to the Discord voice message — it plays at ~0.5x speed

Expected

Voice message plays at normal speed (1x).

Actual

Voice message plays at ~0.5x speed. The audio sounds slowed down.

Analysis

The TTS server outputs MP3 at 24000 Hz sample rate. When OpenClaw converts this to a Discord voice message (opus encoding), it appears to assume 48kHz input, causing the 0.5x playback speed. Sending the same MP3 as a regular file attachment plays correctly.

ffprobe output of TTS source:
Stream #0:0: Audio: mp3 (mp3float), 24000 Hz, mono, fltp, 128 kb/s

Workaround

Resampling the audio to 48kHz before passing to OpenClaw would likely fix it, but this should be handled internally.