Voice dictation on Windows that actually works (in 2026)

Voice dictation on Windows 11 (Win+H) works, but it has three problems that get in the way of serious use:

1. Average accuracy. Proper names, slang, technical terms — it misses often and there’s no easy way to correct.
2. Partial cloud dependency. Some modes send audio to Microsoft’s servers.
3. No per-app hotkey. Only works with Windows’ popup, which steals focus and breaks flow.

If you write a lot — email, Slack, Obsidian, scripts, code — better dictation pays for itself in hours saved per week. Here’s the setup with VoxBooster in 2026.

What changes with local, neural dictation

VoxBooster uses Whisper-derived models running locally on your PC. Practical differences:

• High enough accuracy you don’t need to edit much. Accepts punctuation by command (“comma,” “period,” “new line”).
• 100% local. Nothing goes to a server. Important for anyone dictating sensitive documents, proprietary code or professional conversation.
• Global hotkey. Hold Ctrl+Shift+D in any app — Outlook, VS Code, browser, WhatsApp Web — speak, release, the text appears in the field where the cursor is.

Setup

1. In VoxBooster, go to Dictation and enable.
2. Pick a model: Fast (200 ms latency, good accuracy) or Precise (500 ms latency, high accuracy). For most uses, Fast is enough.
3. Set the primary language and optional secondary — useful if you mix technical English into your speech.
4. Set the global hotkey. Default: Ctrl+Shift+D.
5. Done. Open any app, place the cursor, hold the hotkey, speak, release.

Modes of use

Push-to-dictate. Hold the hotkey, speak, release. Text appears immediately. This is the mode for 80% of cases.

Toggle. Press the hotkey once to turn on, again to turn off. Useful for long dictation (script, big email) where you don’t want to hold a key down.

Always on. Passive recognition waiting for a voice trigger (“Vox, write…”). Useful for hands-free flow like driving or with RSI.

Useful commands

• “comma” → ,
• “period” / “full stop” → .
• “question mark” → ?
• “exclamation mark” → !
• “new line” → line break
• “new paragraph” → two line breaks
• “colon” → :
• “open paren / close paren” → ( / )
• “delete last word” → backspace the word
• “delete everything” → clears the current dictation buffer

Use cases that work well

• Email in volume. Dictating a draft is 3x faster than typing. 30-second review at the end.
• Personal notes in Obsidian / Notion. Capturing an idea at the rhythm it hits your head, not limited by typing speed.
• Replying on Slack / WhatsApp Web. On a long reply, dictating with a global hotkey is faster than switching to phone and using phone dictation.
• Audio messages in text form. Record the idea speaking, get it in text, quick edit. Respects the reader’s time better than audio.
• Scripts, blog posts, documentation. First draft dictated is 2-4x faster than typed. Text edit later, normally.

Cases where dictation doesn’t win

• Code. Variable names, syntax, braces open/close — dictating code is slower than typing with autocomplete. Exception: long comment in a function, which dictates well.
• Tables / spreadsheets. Grid structure doesn’t match voice flow.
• Noisy environment. Room with loud AC, café, coworking — accuracy drops. Better back to typing.

Privacy

The model runs locally, audio never leaves your PC, no dictation log is saved by default. If you want a log — to review lost drafts — enable “Keep history” in settings; files go to a local folder only you can access, encrypted by Windows.

That matters for anyone dictating sensitive conversation. Windows native dictation, even in offline mode, shares device telemetry. VoxBooster doesn’t — no telemetry at all in the dictation module.