Linux

Voice Dictation on Linux: Best Offline Apps and Complete Setup Guide

The best way to get voice dictation on Linux today is a Whisper-based tool: either an open-source script you wire up yourself, or a packaged app that types into any window from one hotkey. This guide compares the real options and walks through setup on Ubuntu, Fedora and Arch, with notes on X11, Wayland and PipeWire. If you want the short version: AirTypes Plain is free and runs offline on your CPU, and Polished cleans up your speech when you want it, at 120 minutes for every dollar.

Why voice dictation on Linux was historically hard

For a long time, Linux users had few real options for voice dictation:

  • Browser voice typing, such as Google Docs voice typing in Chrome: free and decent, but it needs an internet connection and only types inside that one browser tab.
  • Julius and CMU Sphinx: open source, but with dated accuracy and a complex setup.
  • Kaldi-based tools: research-grade engines with no friendly desktop interface.
  • Cloud speech APIs: accurate, but your audio leaves your machine and you have to build the desktop glue yourself.

The well-known commercial dictation apps still skip Linux. Wispr Flow runs on Mac, Windows, iOS and Android. Superwhisper runs on Mac and Windows plus iPhone and Android. SpeechPulse runs on Windows and Apple Silicon Macs. Dragon Professional is Windows software. None of them ships a Linux build at the time of writing.

The desktop environments did not fill the gap either. GNOME and KDE offer accessibility and input features, but nothing gives you the "hold a key, speak, the text appears in any app" experience that people on other systems take for granted.

Skip the DIY setup

AirTypes ships Whisper as a native Linux app (.deb, .rpm and AppImage) with one hotkey for every app. Plain mode is free and runs offline on your CPU. No card.

Download for Linux

How Whisper changed Linux voice dictation

OpenAI's open-source Whisper model changed the equation. Released in 2022 and extended through the large-v3 and large-v3-turbo models, Whisper brought three things Linux dictation had been missing:

  • Strong accuracy on everyday speech, including accented English, from a model you can download and run yourself.
  • CPU-friendly sizes. The model family ranges from tiny (about 78 MB) to large, so even a modest laptop can run one of the smaller sizes without a GPU.
  • Open weights. It can run completely offline, self-hosted, with no licence fee, which is exactly what the Linux community wanted.

Whisper was trained as a multilingual model family, and there are also English-only versions of the smaller sizes that are faster and tuned for English. The result was a wave of Linux dictation tools built on top of it, from developer scripts to packaged apps. For a deeper look at how the engines work, read our offline speech recognition explainer, and for the size trade-offs see Whisper model sizes explained.

Full comparison: best Linux voice dictation tools in 2026

Tool Engine Offline Types into any app Cleanup Setup Price
AirTypes Plain Whisper, on your CPU Yes, 0 bytes uploaded Yes, one hotkey Light: leading um, capital, period Install and go Free
AirTypes Polished Large speech model on AirTypes servers No, needs internet Yes, one hotkey Fillers, false starts, grammar, punctuation Install and go Prepaid credit
OpenWhispr Whisper or Parakeet locally; cloud models optional Yes, with local models Yes, pastes at the cursor Varies by setup App, then models Free
Vocalinux Open-source engines Yes Yes No Manual config Free
Nerd Dictation Vosk Yes Partial (best on X11) No Script-based Free
Speech Note Several, including Whisper and Vosk Yes Mostly its own window No App, then models Free
whisper.cpp + your own script Whisper Yes Manual glue No DIY only Free
Google Docs voice typing Cloud No Browser tab only No Open in Chrome Free

Wispr Flow, Superwhisper, SpeechPulse and Dragon are left out of the table because none of them runs on Linux at the time of writing.

Coming from Wispr Flow, Superwhisper or SpeechPulse?

Many people arrive here after using a dictation app on a Mac or Windows machine and then moving to, or adding, a Linux box. Here is what each one maps to on Linux.

  • Wispr Flow (often spelled Whisper Flow): cloud only, monthly subscription, no Linux app. The closest Linux equivalent is AirTypes Polished for the cleaned-up text, with Plain as the free offline mode Wispr Flow never had. Details in the Wispr Flow alternative guide and the short Whisper Flow alternative for Linux page.
  • Superwhisper: local Whisper models on Mac and Windows, no Linux build. AirTypes Plain is the same idea on Linux, free with no time limit. See the Superwhisper alternative guide.
  • SpeechPulse: local Whisper on Windows and Apple Silicon Macs with a one-time license, no Linux build and no Intel Mac build. AirTypes Plain covers the offline dictation part on Linux without a GPU; the SpeechPulse alternative guide compares them feature by feature.
  • Paying monthly and tired of it: the dictation without a subscription guide works through what a month really costs, and local vs cloud speech to text explains the trade-off behind Plain and Polished.

AirTypes: Plain and Polished on Linux

AirTypes is a desktop dictation app for macOS and Linux. (There is also Dictate in your browser, new in November 2026: Polished only, nothing to install, and the text is copied for you to paste.) The app fills the gap the commercial tools leave on Linux: a maintained app you do not have to wire up by hand. It ships as a native package with Whisper built in, so there is no Python environment to manage. You hold Ctrl + Shift + Space, speak, let go, and the text lands at your cursor in whatever app has focus, from a terminal to a browser to LibreOffice.

There are two modes, and you switch between them on the app's Home screen.

Plain: free, offline, English

  • Whisper runs on your own CPU. 0 bytes of audio are uploaded, and it works in airplane mode or on a machine with networking switched off.
  • Free, with no time limit. There is no clock running down on Plain.
  • English only. Plain uses Whisper's English models.
  • Light cleanup on the machine. It removes a leading um or uh, capitalizes the first letter and adds a period.
  • Two model tiers, Swift and Crisp (see below). Crisp is the tier we recommend.

Here is a real Plain result. Said: "ok so um the meeting is moving to tomorrow at ten, client is not available, and uh send him the invoice again". Typed: "The meeting is moving to tomorrow at ten, client is not available, and uh send him the invoice again." Accurate words, but the mid-sentence "uh" stays.

Polished: cleaned up, in your language

  • Polished sends the recording to AirTypes to transcribe and clean it up. A large speech model on our servers transcribes it, then an AI removes filler words and false starts and fixes grammar and punctuation, keeping your meaning, numbers and names.
  • A guard on numbers and drift. AirTypes checks that the rewrite keeps every number you said and, with translation off, stays close to your length. A rewrite that drifts too far from what you said is dropped, and you get your own words (the raw transcript) instead.
  • Your language. Polished works in English, Hindi, Bangla, Japanese, German and most other spoken languages, so try yours. An optional setting, Always translate to English, lets you speak your language and have English typed. On translations the guard checks numbers only, so check names there.
  • Easy on older laptops. Nothing heavy runs locally, which helps on an old ThinkPad. It does need an internet connection.
  • Prepaid credit. $5 buys 600 minutes of speech, charged by the second for voice and cleanup together. Credit never expires and nothing renews. When credit runs out, Polished pauses and Plain keeps working, free.

The same sentence through Polished comes out as: "The meeting is moving to tomorrow at 10:00. The client is not available. Send him the invoice again." More on the trade-off in how Polished works and on the pricing section.

Plain model tiers

Tier Whisper model Download Suggested hardware
Swifttiny (English)~78 MB4 GB RAM, no GPU
Crispbase (English)~148 MB4 GB RAM, no GPU

Crisp is the tier we recommend, and Swift is the fastest. When you need more accuracy than Crisp, for technical terms, accents or long documents, use Polished: there is no bigger local model to pick.

System requirements (Linux)

  • x86_64 processor with AVX2 (most Intel and AMD chips from about 2015 on), no GPU required: Plain runs on the CPU
  • 4 GB RAM, for either Plain tier
  • Packages for Ubuntu 22.04+, Debian 12+, Linux Mint 21+ (.deb) and Fedora 36+ (.rpm); the AppImage suits Arch and most other x86_64 distributions
  • Microphone: any device your PulseAudio or PipeWire setup can see
  • Display server: an X11 session for the hotkey (on Wayland it may not respond; see the Wayland notes)
  • An internet connection only for sign-in and for Polished; Plain transcription runs without one

You create a free account at first launch. In Plain, AirTypes receives your account email and usage counts, nothing else. If you use Polished, it also receives the recording.

The best free options, honestly assessed

OpenWhispr: open source, on Linux, Mac and Windows

OpenWhispr is a free, MIT-licensed dictation app. A global hotkey dictates into any app and pastes the text at your cursor. It runs local models such as Whisper and NVIDIA Parakeet offline, or cloud models through your own provider account.

On Linux it ships as AppImage, .deb, .rpm and .tar.gz. On Arch, community-maintained AUR packages include openwhispr-bin and openwhispr-appimage. It can use helpers for pasting if you install them: xdotool on X11, wtype or ydotool on Wayland. Details checked on November 7, 2026.

Best for: people who want an open-source desktop app, a choice of local engines, and the same tool on Windows, where the AirTypes desktop app is still in development and AirTypes works only in the browser.

Vocalinux: free open-source Linux voice dictation

Vocalinux is an open-source, offline voice dictation tool for Linux. It types into other applications and works across the common desktop environments. It costs nothing.

Best for: people comfortable with a manual setup who mainly want raw transcription typed at the cursor. A good "it works" solution once configured.

Limitations: manual model management, no cleanup of filler words, and community support only.

Nerd Dictation: lightweight Vosk option

Nerd Dictation is free, open source and scriptable, using Vosk for offline recognition. It is lightweight and good for simple dictation on X11, and you can hook it into your own scripts.

Limitations: command-line setup, lower accuracy than the larger Whisper models, and rougher behaviour on Wayland.

Speech Note

Speech Note is a free, open-source Linux desktop app for offline speech to text. It supports several engines, Whisper and Vosk among them, and you download the models you want from inside the app.

Best for: transcribing notes or longer passages into its own window, then copying them where you need them. It is less suited to the "type into whatever app has focus" workflow.

whisper.cpp, the DIY way

You can run whisper.cpp (or another Whisper port) directly and script transcription yourself. You get Whisper accuracy for free, but you maintain the glue: capturing the microphone, triggering on a hotkey, and sending the text to the focused window, which differs between X11 and Wayland.

Best for: developers who enjoy owning every piece and want full control over models and flags.

Setup on Ubuntu and Debian

AirTypes ships a .deb package for Ubuntu 22.04+ and Debian-based distributions such as Mint and Pop!_OS.

  1. Download the .deb file from the download page.
  2. Install it from a terminal, along with the libxdo3 library it needs in order to start:
    cd ~/Downloads
    sudo apt install ./AirTypes_*.deb libxdo3
  3. Launch AirTypes from your application menu.
  4. Follow the setup: create your free account (email and password, or Google), download a Plain voice model and read the short guide.
  5. Allow microphone access if your desktop asks for it.

AirTypes then lives in your system tray. Hold Ctrl + Shift + Space in any application to start recording.

For Vocalinux or Nerd Dictation, follow the installation guide in each project's repository; expect to install Python packages and audio libraries yourself.

Setup on Fedora

AirTypes ships an .rpm package for Fedora 36+ and openSUSE Tumbleweed.

  1. Download the .rpm file from the download page.
  2. Install it from a terminal, along with the libxdo library it needs in order to start:
    cd ~/Downloads
    sudo dnf install ./AirTypes-*.rpm libxdo
  3. Launch AirTypes, pick your microphone and download a Plain model tier.

Note for Fedora users: Fedora uses PipeWire for audio by default, and your microphone shows up in AirTypes the same way it does in any other app. If it does not, check your input device in the GNOME sound settings first.

Setup on Arch Linux

AirTypes ships an AppImage that works on Arch and other rolling-release distributions.

  1. Download the .AppImage from the download page.
  2. Make it executable:
    chmod +x AirTypes_*.AppImage
  3. Install FUSE, which AppImages need, then run it:
    sudo pacman -S fuse2
    ./AirTypes_*.AppImage
  4. Keep it somewhere stable such as ~/Applications/. For a launcher entry, a tool like AppImageLauncher can add one for you, or you can write a desktop entry yourself.

The AppImage installs nothing else on your system; to remove it, delete the file. For Vocalinux, Nerd Dictation or Speech Note, check each project's page for Arch packages.

Wayland vs Xorg: what to know

Getting text from a dictation app into another app is the part Linux makes tricky, and it depends on your display server. AirTypes gives you three output modes in Settings:

  • Direct Paste (the default): copies the text to the clipboard and presses Ctrl+V for you. Fastest for long text.
  • Type at cursor: types each character at your cursor, which is handy in terminals and code editors.
  • Copy to Clipboard: puts the text on the clipboard and lets you paste it yourself.

Xorg (X11)

X11 lets one application send keystrokes to another, so both Direct Paste and Type at cursor work across your apps. If a terminal emulator uses Ctrl+Shift+V to paste, switch to Type at cursor for that kind of work.

Wayland

AirTypes listens for its hotkey through X11, and Wayland can block that, so on a Wayland session the hotkey may not respond in every app. Ubuntu defaults to Wayland on GNOME, so this matters for most Ubuntu users. Check with echo $XDG_SESSION_TYPE: if it prints wayland, log out, click your name on the login screen, then the gear icon, and choose Ubuntu on Xorg, GNOME on Xorg or Plasma (X11). Some newer desktops no longer offer an X11 session; the Wayland fix on the download page says what to do then.

Wayland's security model also restricts how an external application can send input to other windows, and each compositor (GNOME, KDE Plasma, Sway) draws that line a little differently. If text does not arrive in a particular app, switch to Copy to Clipboard: the transcript is waiting on your clipboard and you paste it with one shortcut.

Tips for best results on Linux

  • Test your microphone first. Run arecord -l to check that your mic is recognized, and use pavucontrol or your desktop's sound settings to check input levels.
  • Check privacy settings. On recent GNOME versions, look under Settings, Privacy, Microphone if an app is not getting audio.
  • Change the hotkey if it clashes. If Ctrl + Shift + Space is taken by your desktop or an input method, set a different one in AirTypes Settings. You can also switch from hold-to-talk to toggle (press once to start, once to stop).
  • Pick the right tier for your CPU. Start with Crisp, the tier we recommend; if it feels slow on an older machine, switch to Swift. If Plain still feels slow, or you want more accuracy than Crisp, Polished does the heavy work off the machine.
  • Start on login. Add AirTypes to your startup applications in your desktop settings, or place a desktop entry in ~/.config/autostart/.
  • Speak in full sentences. Whisper uses context, so a complete sentence comes out better than isolated words.

Final recommendation

The right Linux voice dictation tool depends on what you value:

  • Want install-and-go, free and offline, typing into any app: AirTypes in Plain mode.
  • Want clean, punctuated text or dictation in a language other than English: AirTypes Polished, prepaid at 120 minutes for every dollar, with Plain as the free fallback.
  • Want a free open-source desktop app with a choice of engines: OpenWhispr.
  • Want free open source, fine with manual setup: Vocalinux or Nerd Dictation.
  • Want to transcribe notes into a dedicated window: Speech Note.
  • Want maximum DIY control: whisper.cpp with your own script.

For most Linux users who want something that just works without scripting, AirTypes Plain is the quickest route to offline dictation, and it costs nothing to try. Comparing across platforms? See the best offline speech recognition apps.

FAQ

Does Ubuntu have built-in speech to text?

Not in the way macOS and Windows do. Ubuntu has accessibility and input features, but no built-in, system-wide dictation where you hold a key, speak and the text appears in any app. For that you use a dedicated tool: Nerd Dictation, a whisper.cpp script, or a packaged app like AirTypes.

What is the best free speech-to-text for Linux?

Among open-source tools, OpenWhispr, Vocalinux, Nerd Dictation, Speech Note and whisper.cpp scripts are the usual picks, and most take some manual setup. AirTypes Plain mode is also free: it runs Whisper offline on your CPU, types into any app with one hotkey, and has no time limit. It is English only.

What is the best offline voice recognition for Linux?

A Whisper-based tool. AirTypes packages Whisper into a native Linux app (.deb, .rpm and AppImage). In Plain mode transcription runs on your own CPU, 0 bytes of audio are uploaded, and it works with no internet connection.

Does voice dictation on Linux work on Wayland?

Not reliably. AirTypes listens for its hotkey through X11, and Wayland can block that, so the hotkey may not respond in every app. Run echo $XDG_SESSION_TYPE: if it prints wayland, log in to an X11 session such as Ubuntu on Xorg, GNOME on Xorg or Plasma (X11). Wayland also limits how one app can type into another, which is why AirTypes has three output modes: Direct Paste, Type at cursor, and Copy to Clipboard, which always works because you paste the text yourself.

Do I need a GPU for offline voice dictation on Linux?

No. AirTypes runs Whisper on the CPU. Both Plain tiers, Swift and Crisp, run on a 4 GB machine with no GPU. When you need more accuracy than Crisp, Polished transcribes the recording with a large speech model on the server, so it needs a connection, not stronger hardware.

Related: read how offline speech recognition actually works, check whether voice dictation is private, or start with the complete guide to offline voice-to-text.

Get AirTypes for Linux, free

Plain is free and runs offline on your CPU, with no time limit. Polished is optional prepaid cleanup when you want it. Ships as .deb, .rpm and AppImage for Ubuntu, Fedora, Arch and more. The Windows desktop app is in development: join the waitlist, or Dictate in your browser today.

Download for macOS and Linux