Problem

The OpenAI-compatible TTS provider (used for local TTS servers like Piper/Kokoro) treats responseFormat: "mp3" as not voice-compatible for Telegram voice notes. However, Edge TTS (Microsoft provider) outputs MP3 by default and works fine as Telegram voice messages.

Current behavior

OpenAI provider (speech-provider-CzRYCn2R.js):

voiceCompatible: req.target === "voice-note" && responseFormat === "opus"

When configured with responseFormat: "mp3" for Telegram:

• voiceCompatible = false
• OpenClaw falls back to sendAudio instead of sendVoice
• Telegram displays it as a file attachment with filename, no voice waveform UI

Edge TTS provider (speech-provider-cIDKP9Bn.js):

• Outputs audio-24khz-48kbitrate-mono-mp3
• isVoiceCompatibleAudio() checks file extension (.mp3 is in whitelist)
• Telegram sendVoice accepts MP3 and converts server-side
• Voice note UI (circle + waveform) works correctly

Expected behavior

MP3 from OpenAI-compatible local TTS should behave identically to Edge TTS:

• Pass voiceCompatible check
• Route through sendVoice
• Show native Telegram voice message UI

Workaround

Manually patching the source:

// speech-provider-CzRYCn2R.js line ~189
voiceCompatible: req.target === "voice-note" && (responseFormat === "opus" || responseFormat === "mp3")

Related

• Issue feat(tts): send Telegram voice messages as OGG/Opus (sendVoice) instead of audio files #43072 (feat(tts): send Telegram voice messages as OGG/Opus)
• The architectural fix might be automatic container conversion (WAV/MP3 → OGG/Opus) before sendVoice, rather than relying on Telegram to accept non-Opus formats.

Environment

• OpenClaw version: 2026.5.6
• Channel: Telegram
• Local TTS: Piper TTS (OpenAI-compatible API)
• Server response formats tested: mp3, opus, ogg