Voice Typing on Linux: What Actually Works

If you run Linux, you already know the feeling. The box arrives, you open it, and instead of a finished piece of furniture you get a bag of parts, a folded diagram in six languages, and two extra screws that apparently belong to nothing. It is buildable. It is often better than the pre-assembled version once you are done. But nobody handed you a finished thing. Voice typing on Linux is exactly that flat-pack box.

Windows users press one key and talk. Mac users flip a switch in System Settings and talk. Linux users get the bag of parts. The good news, and the reason this article exists, is that the parts are real, they fit together, and the result can rival the polished versions. You just have to know which pieces to pick up first. Let us assemble it.

The honest truth: Linux has no Win+H

Let us say the uncomfortable part plainly, because pretending otherwise wastes your time. Mainstream Linux desktops do not ship a first-class, system-wide dictation feature the way Windows ships Voice Typing (Win + H) or macOS ships Dictation. Neither GNOME nor KDE Plasma gives you a built-in microphone button that types cleanly into any text field out of the box. There is no single switch to flip.

That is the missing screw. Everything else in this guide is about the parts that genuinely work, so you are not left staring at an empty diagram. The Linux answer is not a product. It is an assembly, and it is a good one.

The open-source parts that actually fit

Here are the pieces worth reaching for, roughly in order of how quickly a normal person can get them working. Each one is genuinely offline, which on Linux is often the whole point.

Speech Note (dsnote): the closest thing to easy

Speech Note is a Flathub app that runs entirely offline and does both speech-to-text and text-to-speech. Its big advantage is model choice: it can run Whisper models and Vosk models locally, so you get modern transcription quality without sending audio anywhere. It ships as a normal graphical app, so if you want to install one thing, click a few times, download a model, and start dictating into its window, this is your least painful starting point. The trade-off is that it is app-first: you dictate into Speech Note and then move the text, rather than talking straight into any field system-wide.

nerd-dictation: system-wide, if you run X11

nerd-dictation is the closest thing to "talk and it types into whatever has focus." It is offline, built on the Vosk engine, and it simulates keystrokes through xdotool, which means it is X11-focused. On a Wayland session (the default on modern GNOME) you will need workarounds or an XWayland path, and that is where the two extra screws show up. It is command-line to set up and hackable by design, so it rewards people comfortable in a terminal. When it clicks, it is the most "real dictation" experience on the list.

whisper.cpp community tools: raw power

OpenAI's Whisper is the same model family that powers a lot of modern dictation, and whisper.cpp is a fast, local, C/C++ implementation that runs it on ordinary hardware. It is not a dictation app by itself, it is an engine, but a healthy ecosystem of community scripts and small front-ends wire it to a hotkey so you can record, transcribe, and paste. If you like assembling your own tools, this gives you the best transcription quality with full local control. Expect to do some wiring yourself.

Talon: powerful, with a learning curve

Talon is a different animal. It is an advanced voice-control system aimed at hands-free computing and accessibility: not just dictation, but commands, cursor control, and full workflows by voice. It is the most capable option here and the one people with RSI or accessibility needs often swear by. The cost is a real learning curve. You are not clicking one button, you are learning a system. If dictation is a means to controlling your whole machine by voice, Talon is worth the climb.

A real setup: dictating with Speech Note in a few minutes

Since Speech Note is the fastest path to your first working dictation on Linux, here is the concrete assembly. This is the version you can finish tonight.

1. 1

   Install from Flathub

   If you have Flatpak set up (most distros do, or see flatpak.org for one-time setup), install Speech Note from its Flathub page. On many desktops you can also find it in your software center by searching "Speech Note."

2. 2

   Download a speech-to-text model

   Open the app, go to its Languages or models section, and download a speech-to-text model for your language. Choose a Whisper model if you want the strongest quality, or a lighter Vosk model if your machine is older. Everything runs locally after the download.

3. 3

   Grant microphone access and test

   Pick your microphone in the app, press the record button, and say a test sentence. The text appears in the Speech Note window. Confirm the language matches what you are speaking before you rely on it.

4. 4

   Move the text where you need it

   Copy the transcribed text and paste it into your editor, chat, or browser field. This is the honest catch: Speech Note transcribes into its own window first, so there is a copy-paste step rather than typing directly into any app.

5. 5

   Optional: wire a global shortcut

   Advanced users bind Speech Note actions or a nerd-dictation toggle to a keyboard shortcut in their desktop settings, so dictation starts with one keypress. That upgrade turns it from a window you visit into something closer to system-wide.

The honest limits (the two extra screws)

Every flat-pack has its awkward bits. Here is where Linux dictation still asks something of you, so you are not surprised later:

• Wayland friction. Tools that type into any window (like nerd-dictation via xdotool) were built around X11. On a Wayland session you may need XWayland or extra configuration to make keystroke injection work.
• Setup is on you. There is no one-click, system-wide switch. Even the easiest option, Speech Note, involves installing an app and downloading a model before you say a word.
• App-first vs system-wide. The friendliest tool (Speech Note) transcribes into its own window; the most system-wide tool (nerd-dictation) is the fussier one to install. You often trade ease for reach.
• Punctuation and names vary by model. Lighter Vosk models tend to skip punctuation and mangle proper nouns. Whisper models handle both far better but cost more compute.
• You maintain it. Community tools move fast and depend on your distro, your session type, and your audio stack. When something breaks after an update, the fix is usually yours to find.

What about Golem? (Windows and Mac only)

You may have seen Golem mentioned as a clean, system-level voice-typing tool: it captures at the OS level, runs OpenAI's Whisper (large-v3-turbo) with an LLM cleanup pass that adds punctuation, supports custom vocabulary for names and technical terms, and can even speak one language and output another. That all sounds relevant here, so let us be blunt about the catch: Golem is Windows and Mac only. It does not run on Linux. On your Linux box it is simply not an option, and no amount of wanting it changes that.

Where it becomes relevant is if you live in more than one operating system. Plenty of Linux users also have a work laptop on Windows, a Mac at home, or a dual-boot partition. On those machines you actually have choices: the built-in Win+H and Apple Dictation (free), Wispr Flow (around $15 per month), Superwhisper (on-device, Mac-focused), and Golem (free to start, then $3 per month or $30 per year). Be clear on Golem's limits before you count on it: it needs a login and an internet connection because the transcription runs server-side, and it only runs on Windows and Mac. But on Linux itself, your best bets are the open tools above, full stop. This is a roundup, not a sales pitch, and on Linux the honest answer points at Speech Note, nerd-dictation, whisper.cpp, and Talon.

Frequently asked questions