Operational notes for TTS provider selection and tuning.

Environment prerequisite
- `edge_tts` can show `ModuleNotFoundError: No module named 'edge_tts'`.
- Fix: `uv tool install edge-tts`. If it still fails, verify `uv` is in PATH and that the harness
  activates the tool environment before invoking `edge-tts` or Python modules it drives.
- List edge voices: `edge-tts --list-voices`.

Provider facts (official Hermes TTS providers)
- Providers are configured under `tts.provider` plus `tts.<provider>` in config.
- Voice toggle in session: `/voice tts | /voice off`.
- Known Hermes TTS providers include: `edge`, `elevenlabs`, `openai`,
  `xai`, `mistral`, `neutts`, and `piper`.

Voice tuning notes (edge TTS)
- The pt-BR catalog in the public Edge TTS is smaller than en-US.
- Verified pt-BR voices in this environment include at least:
  `pt-BR-FranciscaNeural` and `pt-BR-AntonioNeural`.
- Speed is relative: `--rate +70%` gives roughly 1.7x; `--rate +0%` is default.
- Example invocation:
  `edge-tts --voice pt-BR-FranciscaNeural --text "text" --rate +70% --write-media /tmp/out.mp3`

Provider selection order for pt-BR (high to low expected quality)
1. `elevenlabs`: highest naturalness among listed providers; free tier available.
2. `openai`: strong quality, but paid and needs `VOICE_TOOLS_OPENAI_KEY`.
3. `edge`: free and convenient, but limited voice set and lower expressiveness in pt-BR.
4. `neutts` / `piper` local: offline-capable, but generally less natural for pt-BR.

Config references
- Config path: `~/.hermes/config.yaml`, key `tts.provider`.
- Edge voice override: `tts.edge.voice`; example value: `pt-BR-FranciscaNeural`.
- Change provider: `hermes config set tts.provider <provider>`.
- Change edge voice: `hermes config set tts.edge.voice <shortname>`.