hermes - 💡(How to fix) Fix [Feature]: Native Supertonic (on-device ONNX) TTS provider

Problem or Use Case

Hermes ships several native TTS providers (edge, openai, elevenlabs, …) plus local ones (piper, kittentts, neutts), but Supertonic can only be wired in today through the generic tts.providers.<name>: type: command escape hatch — a separate venv and a custom CLI script. There is no first-class way to pick Supertonic from hermes setup tts / hermes tools, no voice/language/speed/quality selection, and no documented config.

Supertonic is a fully local, fast ONNX engine (no torch, ~400MB model, 31 languages, voices M1–M5 / F1–F5) — exactly the class of the existing Piper provider. Users who want a fast, free, multilingual on-device voice currently have to maintain glue code instead of getting it built in.

Proposed Solution

Add Supertonic as a native built-in TTS provider, mirroring the existing Piper integration:

• Runtime in tools/tts_tool.py: lazy _import_supertonic, _generate_supertonic_tts, register in BUILTIN_TTS_PROVIDERS, dispatch branch, WAV→Opus for Telegram voice notes, check_tts_requirements.
• agent/tts_registry.py: add supertonic to _BUILTIN_NAMES.
• Interactive hermes setup tts: pick Supertonic → language → voice (M1–M5 / F1–F5) → speed (0.7–2.0) → quality via total_steps (Fast 5 / Balanced 8 / High 10 / Very High 12).
• hermes tools picker + post-setup installer.
• Config schema tts.supertonic.{voice,lang,speed,total_steps} in hermes_cli/config.py; shipped defaults M1 / en / 1.0 / 8 steps.
• pyproject.toml extra supertonic-tts = ["supertonic>=1.3.1,<2"] (kept out of [all], like the other TTS backends).
• Docs + unit tests mirroring the Piper provider.

Install via pip with lazy-import, no runtime auto-install (explicit setup required), no API keys (fully local).

Alternatives Considered

Keep using the generic type: command provider (the current approach): it works but needs a separate venv + custom script, has no setup UX, no config schema, and isn't discoverable. Rejected because a fast, multilingual local engine in the same class as Piper deserves first-class support.

Also considered shipping it as an external skill, but it's a core TTS backend (like KittenTTS etc.), not a niche integration — so bundling is the right call.

Feature Type

New tool

Scope

Medium (few files, < 300 lines)

Contribution

• I'd like to implement this myself and submit a PR

Debug Report (optional)