Voice (TTS / STT)

Mic dictation and spoken replies run through a Speaches AI server by default, but any compatible STT/TTS provider can be configured via models.toml and services.toml.

To enable voice features:

  1. Add a models entry in models.toml for task = "text-to-speech" and/or task = "speech-to-text".
  2. Add a stt and/or tts block to settings.toml with the speachesUrl or provider-specific settings.

Example notes:

  • models.text-to-speech in settings.toml should reference an [[models]].id that has task = "text-to-speech".
  • Ctrl+T toggles dictation while typing in the CLI.

See config/config.md for the settings.toml format and models.md for declaring TTS/STT models.

Language

English or Magyar, covering the UI and the assistant’s replies, saved as preferences.language in settings.toml and read once at startup; without a saved choice the system locale decides (a Hungarian locale → Magyar, anything else → English).

Voice caveat for Hungarian: dictation needs the multilingual whisper model on the STT server (the English default is an English-only model — point models.speech-to-text at a multilingual entry in models.toml, and set stt.language in settings.toml to override), and spoken replies stay with the configured Kokoro voice (no Hungarian voice) unless models.text-to-speech points at something Hungarian-capable.

Next:

CLI