guide: private dictation on windows 11

Private, offline dictation on Windows 11: what works in 2026.

You want to talk instead of type, and you want your voice to stay on your PC. Windows 11 ships one tool that does that and one that does not, and free open-source apps do it too. This guide takes them in that order and puts Voxmelt, our own app, last. Every outside fact links the page it comes from, with the date we checked it.

the short answer

Four routes, the free ones first

1. Voice access

Built into Windows 11 and offline after a one-time language download. Try it first.

Sources: Use voice access to control your PC & author text with your voice, Microsoft Support; Set up voice access, Microsoft Support. Checked on 24 October 2026.

Read route 1

2. Voice typing (Win+H)

Built in, but online: your audio goes to Microsoft to be transcribed.

Source: Use voice typing to talk instead of type on your PC, Microsoft Support. Checked on 24 October 2026.

Read route 2

3. Open-source apps

Free, can run the speech model on your PC, and work on Windows, macOS and Linux.

Sources: Handy, README, GitHub, cjpais/Handy; OpenWhispr, README, GitHub, OpenWhispr/openwhispr. Checked on 24 October 2026.

Read route 3

4. Voxmelt

Offline on Windows 10 and 11, built for writing. Free tier, then $49 once.

Read route 4
route 1 of 4

Voice access: free, built in, on-device

If you are on Windows 11 and your language is on Microsoft’s list, start here. It costs nothing extra and recognises your speech on the PC.

Microsoft describes voice access as a way to control your PC and author text using your voice. It uses on-device speech recognition and, in Microsoft’s words, works even without the internet.

Source: Use voice access to control your PC & author text with your voice, Microsoft Support. Checked on 24 October 2026.

It is available in Windows 11, version 22H2 and later. The first set-up needs an internet connection, to download the language files that on-device recognition uses for the language you pick.

Sources: Use voice access to control your PC & author text with your voice, Microsoft Support; Set up voice access, Microsoft Support. Checked on 24 October 2026.

The languages and dialects listed today: English (United States, United Kingdom, India, New Zealand, Canada and Australia), Spanish (Spain and Mexico), German, French (France and Canada), Chinese (Simplified and Traditional), Japanese and Italian.

Source: Set up voice access, Microsoft Support. Checked on 24 October 2026.

On Copilot+ PCs it adds fluid dictation, which Microsoft says makes voice-based dictation smoother and smarter. It runs on on-device small language models and is available in all English locales.

Source: Fluid dictation, Microsoft Support. Checked on 24 October 2026.

If you remember Windows Speech Recognition, the older built-in engine: Microsoft lists it as deprecated and no longer being developed, and says it is being replaced with voice access. The deprecation was announced in December 2023. Microsoft’s voice access page says voice access replaces it on Windows 11, version 22H2 and later, from September 2024. Windows Speech Recognition is not on Microsoft’s list of removed features, so Windows 10 still has it.

Sources: Deprecated features in the Windows client, Microsoft Learn; Use voice access to control your PC & author text with your voice, Microsoft Support; Features and functionality removed in Windows client, Microsoft Learn. Checked on 24 October 2026.

route 2 of 4

Voice typing (Win+H): built in, but online

The quickest way to dictate on any Windows PC, and the wrong one if the point is keeping your audio at home.

Select a text field and press the Windows logo key + H. It works on Windows 11 and Windows 10.

Source: Use voice typing to talk instead of type on your PC, Microsoft Support. Checked on 24 October 2026.

It is not offline. Microsoft’s page says that to use voice typing you need to be connected to the internet, and that it uses online speech recognition powered by Azure Speech services. Your audio goes to Microsoft to be turned into text.

Source: Use voice typing to talk instead of type on your PC, Microsoft Support. Checked on 24 October 2026.

To be fair to it: Microsoft’s speech privacy page says voice data is sent to Microsoft only to provide the service and create text transcriptions, that Microsoft does not store, sample or listen to voice recordings without your permission, and that contributing voice clips is optional.

Source: Speech, voice activation, inking, typing, and privacy, Microsoft Support. Checked on 24 October 2026.

On Copilot+ PCs, fluid dictation in voice typing corrects grammar, punctuation and filler words as you speak, using on-device small language models, in all English locales. The same page still says voice typing needs the internet.

Source: Use voice typing to talk instead of type on your PC, Microsoft Support. Checked on 24 October 2026.

If you want Windows to stop sending speech to Microsoft, turn off Online speech recognition under Settings > Privacy & security > Speech. Microsoft notes that only device-based features, such as Narrator, are then available.

Source: Speech, voice activation, inking, typing, and privacy, Microsoft Support. Checked on 24 October 2026.

route 3 of 4

Free open-source apps that run the model on your PC

These two run on Windows, type into other apps and had a release in October 2026. We read each one’s own repository; if your Windows version or language matters, check theirs before you install.

Handy

Free, MIT licensed, offline by design

  • Its README calls it a free, open source and extensible speech-to-text application that works completely offline: press a shortcut, speak, and your words appear in any text field.
  • Models: Whisper (Small, Medium, Turbo and Large) with GPU acceleration when available, and Parakeet V3, which it describes as CPU-optimised.
  • Windows, macOS and Linux. The latest release, v0.9.8 on 3 October 2026, ships Windows installers for x64 and Arm64.

Sources: Handy, README, GitHub, cjpais/Handy; Handy releases, GitHub, cjpais/Handy. Checked on 24 October 2026.

OpenWhispr

Free and MIT licensed, with local or cloud models

  • A voice-to-text dictation app with local models (NVIDIA Parakeet and Whisper) and cloud models: press a hotkey, speak, and your words appear at your cursor.
  • Its README says that with the local models your audio never leaves your device. It also offers cloud processing, so pick a local model if privacy is the reason you are here.
  • macOS, Windows and Linux. The latest release, 1.10.4 on 22 October 2026, ships a Windows installer.

Sources: OpenWhispr, README, GitHub, OpenWhispr/openwhispr; OpenWhispr releases, GitHub, OpenWhispr/openwhispr. Checked on 24 October 2026.

Recordings rather than live dictation? Buzz transcribes and translates audio offline on your computer with OpenAI’s Whisper: audio and video files, plus live audio from the microphone. MIT licensed; version 1.4.5 came out in August 2026. Its README says the Windows app is not signed, so Windows shows a warning when you install it.

Sources: Buzz, README, GitHub, chidiwilliams/buzz; Buzz releases, GitHub, chidiwilliams/buzz. Checked on 24 October 2026.

Open source tells you anyone can read the code, not which model you selected. Whichever app you pick, run the two-minute offline test: Wi-Fi off, dictate a paragraph, and see whether the text still appears.

route 4 of 4

Voxmelt: what it adds, and what it costs

Voxmelt is our app, so weigh this section accordingly. It does the same core job as the apps above, dictation that runs on your own PC, and is built for people who write all day. The free apps may do some of this too; their pages are linked above.

Offline on Windows 10 as well as 11

Voice access needs Windows 11, and Windows 10’s built-in option, Windows Speech Recognition, is no longer being developed. Voxmelt runs on Windows 10 64-bit (build 19041 or later) and Windows 11, and after the one-time model download it dictates with the network cable unplugged.

No graphics card needed to dictate

NVIDIA’s Parakeet v3 runs on your CPU for English plus 24 European languages. Whisper covers roughly 100 languages: tiny on the CPU, every size up to Large v3 Turbo on an NVIDIA GPU.

An accuracy figure you can check

7.8% word error rate with Whisper large-v3 on a GPU and 11.9% with Parakeet v3 on a CPU, across 5 recorded clips, published on 16 July 2026 with the method and every clip. It tested Voxmelt’s own engines, not voice access, voice typing or the apps above.

AI cleanup that stays on the PC

With an NVIDIA GPU of 6 GB or more (Ampere or newer), a local model rewrites and translates what you dictated, on your machine like the dictation.

Voice commands and Mini Mode

Eight built-in voice commands, every phrase remappable: say “go ahead” to start and “wrap up” to stop. Mini Mode shrinks the app to a small always-on-top pill. Pro and Studio add custom vocabulary for the names it gets wrong.

What does leave the PC, listed

Your audio and transcripts never leave it, and the app has no telemetry. What can: a hashed device fingerprint for the trial and licence check, your email and payment records if you buy, and anonymous model and update downloads. A paid licence keeps working offline for 30 days after its last successful check.

The price, as the site shows it today

A free tier with 60 minutes of dictation a day and no card, and a 15-day trial of everything. After that, Pro is $49 and Studio $99, paid once. There is no subscription and nothing to cancel.

What it does not do: control the whole PC by voice, which is voice access’s job, or run on a Mac or Linux, which Handy and OpenWhispr do. Details are on the pricing page and the benchmark; what can leave the PC is listed on the privacy page.

questions

Private dictation on Windows 11, answered

Does Windows 11 have offline dictation built in?

Yes: voice access. Microsoft says it uses on-device speech recognition and works even without the internet, after a first set-up that downloads the language files. It needs Windows 11, version 22H2 or later. Voice typing (Win+H) is the online one.

Sources: Use voice access to control your PC & author text with your voice, Microsoft Support; Set up voice access, Microsoft Support; Use voice typing to talk instead of type on your PC, Microsoft Support. Checked on 24 October 2026.

Does Win+H voice typing work offline?

No. Microsoft’s own page says you need to be connected to the internet to use voice typing, because it uses online speech recognition powered by Azure Speech services.

Source: Use voice typing to talk instead of type on your PC, Microsoft Support. Checked on 24 October 2026.

What happened to Windows Speech Recognition?

Microsoft lists Windows Speech Recognition as deprecated and no longer being developed, and says it is being replaced with voice access, which is available on Windows 11, version 22H2 or later. The deprecation was announced in December 2023, and Microsoft’s voice access page says the replacement applies to Windows 11, version 22H2 and later, from September 2024. Deprecated is not removed: Windows Speech Recognition is not on Microsoft’s list of removed features, and Windows 10 still has it.

Sources: Deprecated features in the Windows client, Microsoft Learn; Use voice access to control your PC & author text with your voice, Microsoft Support; Features and functionality removed in Windows client, Microsoft Learn. Checked on 24 October 2026.

Can I dictate offline on Windows 10?

Yes, with the built-in Windows Speech Recognition, which Windows 10 still has. Microsoft lists it as deprecated and no longer being developed, but not as removed. Microsoft’s speech privacy page says that on Windows 10, with online speech recognition turned off, Windows uses only device-based speech recognition and sends no voice data to Microsoft; it does not list which apps that covers, so dictate with Wi-Fi off to confirm. Voice typing (Win+H) also runs on Windows 10 but needs the internet, and voice access needs Windows 11, version 22H2 or later. For a local app that is still being developed: Handy and OpenWhispr list Windows among their platforms without naming a version, so check their pages for Windows 10, and Voxmelt runs on Windows 10 64-bit, build 19041 or later.

Sources: Deprecated features in the Windows client, Microsoft Learn; Features and functionality removed in Windows client, Microsoft Learn; Speech, voice activation, inking, typing, and privacy, Microsoft Support; Use voice typing to talk instead of type on your PC, Microsoft Support; Use voice access to control your PC & author text with your voice, Microsoft Support; Handy, README, GitHub, cjpais/Handy; OpenWhispr, README, GitHub, OpenWhispr/openwhispr. Checked on 24 October 2026.

Is a free open-source dictation app private enough?

It can be, if the model runs on your PC. Handy says it works completely offline. OpenWhispr offers local and cloud models and says your audio never leaves your device with the local ones, so check which one is selected. Open source lets anyone read the code; the quicker check for you is to turn off Wi-Fi and dictate.

Sources: Handy, README, GitHub, cjpais/Handy; OpenWhispr, README, GitHub, OpenWhispr/openwhispr. Checked on 24 October 2026.

Do I need a graphics card for offline dictation?

Not necessarily. Handy describes its Parakeet V3 model as CPU-optimised and uses GPU acceleration for Whisper when it is available. Voxmelt’s default engine, Parakeet v3, runs on the CPU, and it needs no graphics card for dictation; only its AI cleanup needs an NVIDIA GPU.

Source: Handy, README, GitHub, cjpais/Handy. Checked on 24 October 2026.

How do I check that an app really keeps my audio on the PC?

Turn off Wi-Fi or unplug the cable, then dictate a paragraph. If the text still appears, the recognition runs on your machine; if it stalls or errors, it was running on a server. For a stricter check, watch the app in Windows Resource Monitor while you speak.

Try the free routes, then compare

Voice access costs nothing on Windows 11, and so do the open-source apps. If none of them fits, Voxmelt’s free tier needs no card.

Direct installer, for PCs without the Store; Windows will warn because it is not code-signed.

Read how Voxmelt does offline dictation on Windows.

sources

Where every outside fact comes from

Microsoft’s own pages and each project’s own repository, read on the date given. If one of them has changed since, tell us at help@voxmelt.com and we will fix this guide.

  1. Use voice access to control your PC & author text with your voice, Microsoft Support. Checked on 24 October 2026.
  2. Set up voice access, Microsoft Support. Checked on 24 October 2026.
  3. Fluid dictation, Microsoft Support. Checked on 24 October 2026.
  4. Deprecated features in the Windows client, Microsoft Learn. Checked on 24 October 2026.
  5. Features and functionality removed in Windows client, Microsoft Learn. Checked on 24 October 2026.
  6. Use voice typing to talk instead of type on your PC, Microsoft Support. Checked on 24 October 2026.
  7. Speech, voice activation, inking, typing, and privacy, Microsoft Support. Checked on 24 October 2026.
  8. Handy, README, GitHub, cjpais/Handy. Checked on 24 October 2026.
  9. Handy releases, GitHub, cjpais/Handy. Checked on 24 October 2026.
  10. OpenWhispr, README, GitHub, OpenWhispr/openwhispr. Checked on 24 October 2026.
  11. OpenWhispr releases, GitHub, OpenWhispr/openwhispr. Checked on 24 October 2026.
  12. Buzz, README, GitHub, chidiwilliams/buzz. Checked on 24 October 2026.
  13. Buzz releases, GitHub, chidiwilliams/buzz. Checked on 24 October 2026.

Voxmelt’s figures come from our benchmark, published 16 July 2026, the system requirements, the privacy page and the pricing page.