Voice translation
Choose an engine and languages under Settings → Text processing → Voice translation. Its shortcut is configured separately and defaults to Right Control.
Choose a translation engine before first use. Until then, the original transcript is kept and no translation request is sent. Existing engine selections are preserved.
Translation engines
Section titled “Translation engines”| Engine | Preparation & network | OS requirement |
|---|---|---|
| Apple native translation | No API key; uses installed on-device language resources. Downloading system resources may need a network connection | macOS 26+; select it manually |
| Large language model | Uses the cloud, local, or compatible model configured in Text processing; network needs depend on the provider | The app’s supported macOS 14+ versions; select it manually and configure a model |
Apple’s supported language pairs depend on the system. If a language pack is missing, follow the error’s instructions to prepare translation resources in System Settings, then retry, or select a configured LLM. A target appearing in the menu does not guarantee Apple supports that pair.
The language-pack check reports resource availability without trying a translation. It does not guarantee success or a fixed latency. On macOS 26.4+, native translation prefers installed low-latency packs, then tries an installed Apple Intelligence high-fidelity model if unavailable. Download the languages you use to avoid relying only on high-fidelity inference, which Apple Intelligence may refuse under critical memory pressure. macOS 26.0–26.3 uses traditional packs.
Modes & languages
Section titled “Modes & languages”| Mode | Behavior |
|---|---|
| Fixed target | Translate the transcript into a chosen language |
| Bidirectional | Translate between A and B according to the input language, such as English and Chinese |
The selector offers 14 presets: Simplified Chinese, Traditional Chinese, English, Japanese, Korean, Spanish, German, French, Portuguese, Russian, Italian, Vietnamese, Thai, and Indonesian. Results and supported language pairs depend on the engine.
Local Only & fallback
Section titled “Local Only & fallback”- Apple native translation and prepared local LLMs can run under Local Only. Prepare models and system language resources first; initial downloads are separate from offline inference.
- Public LLM requests are blocked; allowed loopback endpoints can still run.
- When translation is unconfigured, fails, or is blocked, the app shows a reason and falls back to the original transcript for output.
- A selected Apple engine never switches automatically to an LLM or sends a cloud translation request, even on an unsupported system or after a failure. Choose another engine manually if needed.
Translation follows your output action and local text-processing settings. Esc during translation attempts to output the original transcript when available. See Shortcuts & recording.