Voice Reply

**Judgement:** A solid, privacy-friendly local TTS skill that is well-suited for Telegram voice replies in German and English, provided the host environment is correctly configured. **What it does well** - Converts arbitrary text to speech entirely offline using Piper via sherpa-onnx. - Auto-detects language (German vs English) and picks an appropriate voice. - Emits Telegram-compatible voice notes with the `[[audio_as_voice]]` tag and `MEDIA:` path, making integration straightforward for voice bubbles. - Avoids cloud costs and latency; once installed, it should be fast and responsive on typical hardware. **Main risks / limitations** - **Environment complexity:** Requires sherpa-onnx, Piper voice models, and ffmpeg installed in specific locations with environment variables properly set; misconfiguration will cause failures. - **Platform assumptions:** Documentation targets Linux (x64); other platforms may need manual adaptation and could be less reliable. - **Limited out-of-the-box voices:** Only one German and one English voice are included by default; adding more voices requires manual script changes. - **Language scope:** Auto-detection is designed around German vs English; content in other languages will likely fall back to English or behave suboptimally. **Recommended scenarios** - Bots or assistants running on self-hosted servers where privacy and offline capability are important. - Telegram-based assistants that need to respond with voice notes when users ask to "hear" the answer or request a spoken reply. - Workflows where network access is unreliable or restricted, but TTS is still required. - Privacy-sensitive deployments (e.g., internal tools) that must avoid sending text to external TTS providers.