Prepare your Mac before going offline
Use Voiced’s local mode to transcribe speech directly on your Mac when writing private notes, handling sensitive work messages, or entering text without internet access.
Before traveling or disconnecting, check these requirements while online:
- An Apple silicon Mac with macOS 26 or later: The Voiced app supports macOS 14 or later, but on-device speech recognition requires Apple silicon and macOS 26 or later.
- Sign-in and permissions completed: Sign in to your Voiced account while online and follow the setup prompts to grant microphone access and other required permissions.
- The appropriate Apple speech model downloaded: Your Mac must support your selected primary language, and its language model must be downloaded while online.
- One primary language selected: Local mode transcribes using your selected primary language and custom vocabulary. For example, choose Chinese (中文) or English (英文) in the interface. The menu also offers other languages; whether they work offline depends on Apple’s on-device support and whether the model is ready.
Local mode focuses on transcribing speech directly on your Mac. For Cantonese, Taiwanese, mixed Chinese–English or Cantonese–English dictation, the four built-in editing levels—Original, Cleanup, Fluent, and Professional—translation into 11 languages, or Chinese-variant conversion, use cloud recognition with the press-once-to-start, press-again-to-finish workflow. These cloud features are available on all plans within their respective cloud allowances. Custom-style instructions are exclusive to Pro.
Enable and configure on-device recognition
Once your Mac is ready, follow these steps:
- Open Settings → Language → Primary Language (設定 → 語言 → 主要語言) and choose the language for on-device recognition, such as Chinese or English.
- Go to Settings → Data & Privacy → Local Mode (設定 → 資料與隱私 → 本地模式) and turn on Enable Local Mode (啟用本地模式).
- If you see Prepare Model (準備模型), click it while online and wait until the model is ready.
- To always use on-device transcription in a particular app, select that app in Apps (各 App), expand Advanced → Recognition Mode (進階 → 辨識模式), and change Cloud (雲端) to Private Local (隱私本地).
For example, set Notes or a private journaling app to Private Local, while leaving your email or international messaging apps on Cloud for editing and translation. Each app’s Recognition Mode works independently, whether or not Use app-specific style (使用專屬風格) is enabled. See Per-app settings to configure tone and translation separately for different apps.
Switch to local mode automatically when offline
If you usually use cloud recognition but want to keep dictating when you lose internet access, enable Automatically Switch to Local Mode When Offline (無網路時自動改用本地模式) under Settings → Data & Privacy → Local Mode.
Voiced checks this when recording starts. If Cloud is selected, there is no internet connection when you start recording, and the on-device speech model is ready, Voiced automatically uses local mode to transcribe that recording.
Keep these details in mind when using automatic fallback and local recording:
- Losing the connection mid-recording does not trigger a switch: If you are online when recording starts but lose the connection partway through, that cloud recording will not switch to on-device recognition midway.
- Reconnecting does not automatically upload recordings: Recordings made in local mode stay on your Mac and are not uploaded just because your internet connection returns.
- You can manually request cloud re-recognition: If you have saved a local recording and later want cloud recognition or editing, manually request cloud re-recognition for that history entry. Voiced first asks you to confirm the audio upload, with an option to remember your confirmation choice. This uses your plan’s cloud allowance.
Manage Screen Context and Smart Personalization
For cloud recognition, you can control screen-text access and personalized learning separately to suit your work and privacy preferences. Both per-app controls work independently of Use app-specific style.
Screen Context: off by default
When Screen Context (螢幕語境) is on, Voiced uses the macOS Accessibility permission to read a limited amount of text from the target window—for example, its title, the field where the cursor is currently placed, and visible text. It reads text rather than taking screenshots, and provides that text as reference for cloud processing. With Private Local recognition, Screen Context reading is skipped entirely.
- Global switch: Find it under Settings → General → Screen Context (設定 → 一般 → 螢幕語境).
- Exclude a particular app: Go to Apps → Advanced → Recognition & Personalization (各 App → 進階 → 辨識與個人化) and turn off Screen Context for that app. This exclusion still applies when the global switch is on or the app does not use an app-specific style. Keep the app in the Apps list: removing it also removes its saved exclusion setting.
Smart Personalization: a Pro feature
Smart Personalization (智慧個人化) requires Pro, with both learning and history retention enabled. It learns frequently mentioned vocabulary and topics from your dictation history to help with future recognition.
If you turn off Smart Personalization for an app under Apps → Advanced → Recognition & Personalization, Voiced stops learning from dictation in that app and does not apply previously learned Smart Personalization content there. Custom instructions, domain information, and background information that you entered manually keep their existing settings.
Set local history retention and understand cloud data handling
Choose how long dictation history is stored on this Mac:
- Go to Settings → Data & Privacy (設定 → 資料與隱私).
- Set separate retention periods for Transcript Retention (轉錄紀錄保留) and Original Recording Retention (原始錄音保留).
- If you set Transcript Retention to Never Save (完全不存), new dictations will no longer save either transcript text or original audio on your Mac. You can also manually clear existing local history on the same page.
Voiced applies specific protections to cloud processing and data retention, with different rules for different types of data:
- In ordinary cloud recognition and text-processing workflows, audio is released from the cloud as soon as speech recognition is complete—zero retention—and is not used for model training or fine-tuning. Ordinary transcript text and personalization reference text are also released when the request is complete.
- If input text is flagged for suspected illegal activity or abuse and further investigation is needed, the text-processing provider may retain the relevant text for necessary safety review for up to 90 days. Authorized personnel may review it during that period. This exception does not include audio, and the retained content is still not used for model training or fine-tuning. Text potentially subject to this review includes the transcript, custom instructions, settings or Screen Context text, and personalization text actually supplied in that request.
- History stored locally on your Mac, feedback you choose to submit, and account, usage, payment, and security records are each handled under their own retention rules.
For full terms and provider details, see the Privacy Policy and Subprocessors. See Plans and pricing for cloud allowances across all plans and Pro features. To combine cloud text editing with custom instructions, read Cleanup and rewriting.
Open Settings → Data & Privacy in Voiced, enable local mode, and prepare your speech model so offline dictation is ready whenever you need it.