Voice & language

Voice is a work in progress 🐞 Expect rough edges.

Voice

Mic dictation and spoken replies work in local mode, by default through a Speaches AI server. Any compatible STT/TTS provider works: the models go in models.toml, the endpoint in settings.toml. Ticking Speaches in the setup wizard writes both.

Run a Speaches server (or an equivalent).

Declare the models in models.toml:

[providers.speaches]
base_url = "http://localhost:8000"

[models.kokoro-82m-v1-0-onnx-fp16]
model = "speaches-ai/Kokoro-82M-v1.0-ONNX-fp16"
provider = "speaches"
tasks = ["tts"]

[models.faster-distil-whisper-small-en]
model = "Systran/faster-distil-whisper-small.en"
provider = "speaches"
tasks = ["stt"]

Point settings.toml at the server:
```
[stt]
speachesUrl = "ws://localhost:8000"
language = "en"

[tts]
speachesUrl = "http://localhost:8000"
voice = "af_heart"
```
Dictation uses the realtime WebSocket API (ws://) and spoken replies use plain HTTP, hence the two schemes.

Then Ctrl+T toggles dictation while you type, and preferences.voice = true reads replies aloud.

Language

The interface and the assistant’s replies follow preferences.locale:

Code	Language
`en-GB`	British English
`en-US`	American English
`hu-HU`	Magyar
`nan-TW`	臺語 (Taiwanese Hokkien)
`zh-TW`	繁體中文

The wizard asks first. With no saved value, your system locale decides (LC_ALL, LC_MESSAGES or LANG), and anything unsupported becomes British English. In cloud mode your account’s language is used, and the web app offers the same five.

Voice lags behind:

Dictation needs a multilingual Whisper model. The default above is English-only, so put a multilingual one that lists stt above it and set stt.language.
Spoken replies use the configured Kokoro voice, which has no Hungarian. Put a tts model that does above it.

Abilities