tts-setup

# TTS Setup Configure Text-to-Speech for Hermes Agent. Covers provider selection, voice discovery, language-specific voice lists, and config patching. ## Providers | Provider | Free? | Env Var | Best For | |----------|-------|---------|----------| | **Edge TTS** | Yes (default) | None | Free, many languages/voices | | **ElevenLabs** | Free tier | `ELEVENLABS_API_KEY` | Highest quality, voice cloning | | **OpenAI** | Paid | `VOICE_TOOLS_OPENAI_KEY` | Natural voices (alloy, echo, fable, onyx, nova, shimmer) | | **NeuTTS** | Free (local) | None (`pip install neutts[all]` + `espeak-ng`) | Offline, voice cloning via reference audio | | **Piper** | Free (local) | None | Lightweight, fast, offline | | **Mistral** | Paid | `MISTRAL_API_KEY` | | | **xAI** | Paid | `XAI_API_KEY` | | ## Config Location ``` ~/.hermes/config.yaml # Standard install /hermes-home/config.yaml # When HERMES_HOME=/hermes-home ``` Find it with: `hermes config path` or check `get_hermes_home()` in Python. ### TTS Config Structure ```yaml tts: provider: edge # edge | elevenlabs | openai | xai | minimax | mistral | gemini | neutts | kittentts | piper edge: voice: es-ES-AlvaroNeural elevenlabs: voice_id: pNInz6obpgDQGcFmaJgB # Adam model_id: eleven_multilingual_v2 openai: model: gpt-4o-mini-tts voice: alloy neutts: ref_audio: "" ref_text: "" model: neuphonic/neutts-air-q4-gguf device: cpu piper: voice: en_US-lessac-medium ``` ## Voice Discovery Edge TTS voices are discovered via the Hermes venv Python: ```bash # List all voices /opt/hermes/.venv/bin/python -c " import asyncio, edge_tts async def main(): voices = await edge_tts.list_voices() for v in sorted(voices, key=lambda x: x['ShortName']): print(f\"{v['ShortName']:40s} | {v['Gender']:8s} | {v['FriendlyName']}\") asyncio.run(main()) " # Filter by language (e.g., Spanish) /opt/hermes/.venv/bin/python -c " import asyncio, edge_tts async def main(): voices = await edge_tts.list_voices() es = [v for v in voices if v['Locale'].startswith('es-')] for v in sorted(es, key=lambda x: (x['Gender'], x['ShortName'])): print(f\"{v['ShortName']:40s} | {v['Gender']:8s} | {v['FriendlyName']}\") asyncio.run(main()) " ``` If `edge_tts` is not installed in the venv: ```bash uv pip install edge-tts --system -q ``` ## Patching the Voice ```bash # Find the current voice grep -A 3 "^tts:" /hermes-home/config.yaml | head -5 # Patch it (use the patch tool, not sed) # Old: voice: en-US-AriaNeural # New: voice: es-ES-AlvaroNeural ``` ## Linked Files - `references/spanish-voices.md` — Complete list of all 45 Spanish (es-*) Edge TTS voices ## Pitfalls - **Default voice is `en-US-AriaNeural`** (female, English). Always check what's configured before assuming. - **Edge TTS CLI (`edge-tts`) may not be in PATH** — use the venv Python directly: `/opt/hermes/.venv/bin/python` - **`edge_tts.list_voices()` is async** — must be called with `await` inside an `asyncio.run()` context. - **Voice names are case-sensitive** — use exact `ShortName` from the list (e.g., `es-ES-AlvaroNeural`, not `alvaro`). - **Config changes require `/reset` or gateway restart** to take effect.