Best AI voice typing apps for Linux (2026): 16 that actually run
Written by Futsy, founder at Vibe Typer
As of October 2026, the best voice to text software for a Linux desktop is Vibe Typer: no setup, native on Wayland and X11, and Magic Formatter cleans up what you say before it lands. Sixteen tools ship a real Linux build; fifteen well-known ones, including Wispr Flow and Dragon, do not, and Superwhisper only has early builds for Arch-based desktops.
Most "best voice to text software" lists are written for Windows and Mac, then have Linux bolted on by someone who has never run it. This one is the other way round. Every tool below ships a real Linux build, and the column that matters most is whether it works under Wayland, because that is where most of them quietly fall over.
We make one of these tools, so weigh the recommendations accordingly.
TL;DR
- Want it to just work, no setup → Vibe Typer. One AppImage, native Wayland and X11, cloud-based.
- On Wayland and short on patience → Vibe Typer. It types natively on Wayland, with no
ydotoolsetup. - Write in several languages → Vibe Typer. 99 languages, automatic detection, no downloads.
- Dictate in VS Code and everywhere else → Vibe Typer. VS Code's built-in dictation stops at the window edge.
Why Linux dictation is harder than it should be
Dictation has two halves: turning speech into text, and getting that text into the app you are looking at. The first half is solved. Whisper made good transcription free. The second half is where Linux tools break.
On X11, any process can synthesise keystrokes into any window. That is what xdotool does, and it is why nearly every Linux dictation tool built before about 2022 assumed X11 and worked fine.
Wayland deliberately removed that. A Wayland client cannot inject input into another client; that was the entire point of the security model. So tools built on xdotool or XTestFakeKeyEvent do not degrade gracefully; they do nothing at all, usually without an error message. The workarounds are ydotool, which needs a uinput daemon and often a reboot, dotool, or wtype. wtype does not work on GNOME, since GNOME does not implement the virtual-keyboard protocol.
That is why the honest column below is "Wayland," not "Linux." Ubuntu has defaulted to Wayland since 22.04 and Fedora since 25. If a tool advertises "Linux support" and means X11, a distro upgrade can silently break your dictation.
We wrote up how to test any tool for this in about five minutes: voice typing on Wayland, and why most tools break.
The comparison
Vibe Typer transcribes in the cloud, on large models and server hardware, so the speed and accuracy you get are the same on any laptop. A tool that runs a local model depends on the machine: the large Whisper models that transcribe best are slow on a laptop CPU, and the small ones that keep up miss more.
| Tool | Wayland | X11 | Install | Licence | Cost |
|---|---|---|---|---|---|
| Vibe Typer | Native | Native | AppImage | Proprietary | Free tier, Pro $8 a month billed annually |
| Vocalinux | Yes | Yes | One-line script | AGPL-3.0 | Free |
| VOXD | Yes, via ydotool | Yes | .deb / .rpm / .pkg | MIT | Free |
| Handy | Partial | Yes | AppImage | MIT | Free |
| Myna (preview) | GNOME only | No | PPA | AGPL-3.0 | Free |
| nerd-dictation | Partial | Yes | pip + git + model | GPL-3.0 | Free |
| Talon | No | Yes | Installer / zip | Proprietary + Patreon | Free |
| OpenWhispr | Partial | Yes | Electron app | Open source | Free |
| Spokenly | Yes | Yes | .deb / .rpm / AppImage | Proprietary | Free with local models, Pro $99.99/yr |
| Whispering | Partial | Yes | Desktop app | Open source | Free |
| Speech Note | Standalone | Standalone | Flatpak | Open source | Free |
| whisper.cpp | CLI only | CLI only | Build from source | MIT | Free |
| Blahst | Partial | Yes | Script | Open source | Free |
| Numen | Yes | n/a | Source | Open source | Free |
| VoiceKey | Partial | Yes | .deb | Proprietary | Paid |
| Browser dictation | n/a | n/a | None | n/a | Free |
"Standalone" means it transcribes into its own window rather than typing into whatever app you are using. "Partial" means it works but needs a helper (ydotool, dotool, wtype) and may misbehave on some compositors.
The tools
Vibe Typer
Ours. One portable AppImage, with no repository, no dependencies and no models to download:
chmod +x ~/Downloads/VibeTyper.AppImage
~/Downloads/VibeTyper.AppImage
The reason it sits at the top of that table is the Wayland column. It implements the Wayland and X11 paths natively rather than shelling out to xdotool, picks the right one for your session, and migrates your settings across if you switch. That is unusual, and it is the most common reason people arrive here after something else stopped working.
Press a hotkey, talk, and text lands where your cursor is: terminals, editors, browsers, chat, email. Magic Formatter strips the ums, false starts and self-corrections on the way through rather than transcribing them faithfully, and custom instructions tell it how you like your text every time. Select any text and say "make this shorter" to rewrite it in place. The dictionary uses smart matching and offers to learn from your corrections, paste behaviour is adjustable per app, and 99 languages work with automatic detection and no per-language downloads.
It is cloud-based, and that is a choice rather than a gap. Transcription runs on large speech models and inference hardware built for speed, not on your CPU. That is why it is as fast and accurate on a five-year-old ThinkPad as on a workstation, and why there is no model to download. Audio is processed in memory and discarded, the backend keeps no server-side transcript cache at all, and nothing trains on your voice.
Free plan is 2,000 words and 20 AI operations a month, no card. Pro is $8/month billed annually, $10 month-to-month.
Best for: people who want dictation working in every app on day one and do not want to maintain it.
Vocalinux
The most polished offline option, built for Linux from the start. One command, and the installer detects your hardware:
curl -fsSL https://raw.githubusercontent.com/jatinkrmalik/vocalinux/main/install.sh -o /tmp/vl.sh && bash /tmp/vl.sh --interactive
Tested on Ubuntu 22.04+, Debian 11+, Fedora 39+, Arch, openSUSE Tumbleweed, Manjaro and EndeavourOS. Works on both X11 and Wayland, which puts it in a very short list. Defaults to whisper.cpp with Vulkan for multi-vendor GPU support, and can also use Whisper, VOSK, or a remote API you control. No cloud, no telemetry, no account. AGPL-3.0, no premium tier.
VOXD
Local dictation that types straight into whatever window has focus, shipped as real distro packages rather than a script:
# Ubuntu / Debian
sudo dpkg -i voxd_*.deb
# Fedora
sudo rpm -i voxd-*.rpm
# Arch
sudo pacman -U voxd-*.pkg.tar.zst
whisper.cpp underneath with GGML models from 75MB to 2.9GB, 99+ languages, no GPU required. The author reports testing on Arch/Hyprland, Omarchy 3.0, Ubuntu 24.04 and 25.04, Fedora 42, Pop!_OS 22, Mint 22 and openSUSE Leap 15.6. Optional AI post-processing through llama.cpp or Ollama locally, or a cloud provider if you want one. MIT.
Wayland works but needs ydotool and a reboot; the setup script handles that for you.
Handy
The friendliest way into local Whisper. AppImage for Linux plus macOS and Windows builds, MIT licensed, actively developed.
Two model families. Whisper covers Small through Large and is GPU-accelerated where available. Parakeet V3 is CPU-optimised, auto-detects language, and runs around 5× real-time on mid-range hardware. That second option is the interesting one. It means usable offline dictation on a laptop with no discrete GPU.
Wayland is the weak spot. You will need wtype or dotool, you have to bind the hotkey in your desktop environment rather than in Handy, and the project's own documentation warns that some compositors treat Handy's overlay as the active window, which can land the pasted text in the wrong place.
Myna
sudo add-apt-repository ppa:canonical-desktop-team/myna
sudo apt install myna-config
Canonical's own speech to text tool for Ubuntu Desktop, released as a technology preview on 12 October 2026. Press a shortcut, speak, press it again, and the text lands in the focused app. It inserts text as an input method rather than simulating keystrokes, so it works on GNOME's Wayland session without ydotool. You choose the model (Parakeet, Whisper or FunASR), with GPU engines for NVIDIA cards, and audio stays on the machine.
The limits are those of a preview: Ubuntu 24.04, 26.04 and 26.10 only, GNOME on Wayland only (no X11 sessions or other desktops yet), no published accuracy figures, and no cleanup beyond the model's own punctuation.
Compared with Vibe Typer in the Ubuntu guide.
nerd-dictation
The one everyone names first in Linux dictation threads, and still the most hackable. VOSK-backed, entirely offline, no daemon, and scriptable to a degree nothing else here matches. You can rewrite the output in Python before it is typed.
pip3 install vosk
git clone https://github.com/ideasman42/nerd-dictation.git
cd nerd-dictation
wget https://alphacephei.com/kaldi/models/vosk-model-small-en-us-0.15.zip
unzip vosk-model-small-en-us-0.15.zip && mv vosk-model-small-en-us-0.15 model
Needs Python 3.6+, an audio utility (parec, sox or pw-cat) and an input simulator. That last choice is the Wayland question: xdotool is the default and X11-only, while ydotool, dotool and wtype cover Wayland with the usual caveats.
Be honest with yourself about the trade. VOSK's small models are noticeably behind Whisper on accents and background noise, and there is no AI cleanup, so you get a faithful transcript, ums included. What you get back is total control and no dependency on anyone else's service. GPL-3.0.
Full breakdown of the trade in nerd-dictation vs Vibe Typer.
Talon Voice
The only tool here that does what Dragon does. Dictation is a small part of it. It does full voice control, noise control so you can click with a hiss or a pop, eye tracking for cursor position, and a Python-scriptable command grammar. The community command sets for editors and IDEs are mature.
Talon's Linux support is X11 only. If you are on Wayland this is not currently your tool, and it is the most-requested thing in its community.
Free, with an optional Patreon for early access and priority support. Steep learning curve. You are learning a command language, not installing an app.
OpenWhispr
An Electron desktop app that treats Linux as a first-class target, running Whisper and NVIDIA Parakeet entirely on your machine with no internet required. Open source, 99+ languages, system-wide hotkey.
Spokenly
Ships deb, rpm and AppImage builds (Ubuntu 24.04, Fedora 40 and Debian 13 or later for the native packages, Ubuntu 20.04 or later for the AppImage) and states support for both X11 and Wayland, with the caveat that shortcuts and text insertion vary by compositor. Free with local Whisper and Parakeet models or your own API keys, no account needed. Pro, which uses Spokenly's cloud transcription, is $99.99 a year. Also on Windows, macOS and iPhone.
Whispering
Open source, local Whisper, cross-platform, with one of the more approachable interfaces in the category.
Speech Note
flatpak install flathub net.mkiol.SpeechNote
Fully offline speech-to-text, text-to-speech and machine translation, with strong multilingual coverage: Catalan, Czech, German, Danish, Spanish, Persian, French, Icelandic, Italian, Korean, Dutch, Polish, Portuguese, Tamil and Ukrainian among others. Optional GPU add-ons for AMD and NVIDIA.
One thing to be clear about: it is a standalone notes app, not a system-wide typing tool. You dictate into Speech Note and copy the text out. If the goal was to talk into Slack, this is the wrong shape.
whisper.cpp
Not a dictation app, but a fast C++ inference engine for Whisper models, and the foundation several tools above are built on. Building dictation on it means writing your own hotkey handling, audio capture and text injection. MIT. We walk the whole pipeline in Whisper dictation on Linux.
Blahst, Numen, VoiceKey
Three smaller projects worth knowing about. Blahst sits on top of whisper.cpp to give you speech input in any field. Numen is voice control rather than dictation, designed for Wayland from the start, a lighter-weight answer to the problem Talon solves on X11. VoiceKey, formerly VoiceHotKey, is a paid .deb for Ubuntu, Debian and Mint.
Browser dictation
Chrome and Google Docs have had voice typing for years. It only works in a Chrome tab: not your terminal, not your editor, not Slack desktop, not your email client. Not something you can use all day.
VS Code built-in dictation
Since VS Code 1.131 (July 2026) the editor ships its own on-device dictation for the chat box, the editor and the integrated terminal, on Linux x64 and Arm64 with glibc 2.34 or later. It runs offline after the first model download and works in Remote SSH and Dev Container sessions. Like browser dictation, it is scoped to its own window: nothing outside VS Code gets typed, and its terminal mode strips punctuation, which is wrong for prose prompts to a CLI agent. Details in voice dictation in VS Code and the Linux terminal.
Who should choose what
"I want to stop typing so much and I don't want a project." Vibe Typer. Free tier, one file, works on both display protocols on day one.
"I'm on Wayland and tired of things silently not working."
Vibe Typer. It types natively on Wayland and X11 and picks the right path for your session, so there is no ydotool setup to assemble first.
"I write in several languages." Vibe Typer, with 99 languages, automatic detection and no downloads. Most VOSK-based tools need a separate model per language.
"I mostly dictate prompts to coding assistants." Vibe Typer, whose per-app paste is safe in terminals. See Claude Code, Cursor, OpenCode and dictating over SSH, because remote sessions are where most setups fall down.
"I'm on a specific distro." Ubuntu and Linux Mint have their own guides.
The tools that don't run on Linux at all
Worth recording, because most "best dictation app" roundups list these without mentioning that they skip Linux. Checked against the platform support each vendor publishes:
| Tool | Platforms the vendor lists | Source |
|---|---|---|
| Wispr Flow | macOS, Windows, iPhone, Android; nothing for Linux (an unofficial community port repackages the Windows app) | wisprflow.ai |
| Willow Voice | macOS, Windows (added January 2026), iPhone | willowvoice.com |
| Aqua Voice | macOS, Windows, iPhone, Android | aquavoice.com |
| Dragon Professional | Windows only; Wine and VMs explicitly unsupported | dragon.nuance.com |
| TypeWhisper | macOS, Windows and iOS stable; Android beta | typewhisper.com |
| Soniox Voice Typing | macOS, Windows, iOS, Android | soniox.com |
| SpeechPulse | Windows, macOS (Apple silicon) | speechpulse.com |
| PulseScribe | macOS, Windows; Linux on roadmap | pulsescribe.me |
| Talkativ | Windows | talkativ.app |
| DictaType | Windows | dictatype.com |
| BetterDictation | macOS | betterdictation.com |
| Whispur | macOS | whispur.app |
| EdgeWhisper | macOS (Apple silicon) | edgewhisper.com |
| VoiceInk | macOS 14.4+ | VoiceInk |
| FluidVoice | macOS (iOS and Windows announced) | GitHub |
Superwhisper sits between the two lists. Since October 2026 it has published early Linux builds for Omarchy and other Arch-based desktops, but nothing packaged for Ubuntu, Fedora, Debian or Mint.
Migrating from one of these? Wispr Flow, Superwhisper, Willow Voice, Dragon and TypeWhisper have dedicated guides.
Installing Vibe Typer
- Download the AppImage from the downloads page.
chmod +x ~/Downloads/VibeTyper.AppImage- Run it and grant microphone access.
- Set a global hotkey.
- Click into any app and talk.
If a particular terminal handles inserted text oddly, see text insertion and paste preferences. Full overview at voice typing for Linux.
FAQ
What is the best voice to text software for Linux?
Vibe Typer, for most Linux desktops: no setup, native on Wayland and X11, and Magic Formatter cleans up the ums and self-corrections. The big names in voice to text, Wispr Flow, Willow Voice, Aqua Voice and Dragon, do not ship Linux dictation apps, and Superwhisper only has early builds for Arch-based desktops; the table above only lists tools that do.
What is the best dictation software for Linux?
Vibe Typer, for most people: no setup, it works on Wayland and X11 immediately, and Magic Formatter strips the ums and self-corrections, so clean text lands in any app, terminals included. For ten of them side by side, see Linux dictation software compared.
Is there free dictation software for Linux?
Yes. Vibe Typer is free to start with no card: 2,000 words and 20 AI operations a month, with Magic Formatter on the free plan too. Vocalinux, VOXD, Handy, nerd-dictation, Whispering and Speech Note are free and open source.
What Linux dictation software works on Wayland?
Vibe Typer works natively on Wayland, GNOME included, with no extra setup, and picks the right path for your session. Vocalinux and VOXD also work without extra assembly. Handy, nerd-dictation, OpenWhispr and Whispering can work with ydotool, dotool or wtype, with caveats depending on your compositor. Talon is X11 only.
What happens to my audio in Vibe Typer?
Vibe Typer transcribes in the cloud. Audio is processed in memory and not stored, the backend keeps no transcript cache, and nothing trains on your voice.
Does dictation work in the terminal on Linux?
With a system-wide tool, yes: text goes wherever your cursor is, including terminals and editors. Browser-based dictation cannot do this. For remote work see dictating over SSH.
Which is more accurate, Whisper or VOSK?
Whisper, clearly, especially on accents and background noise. VOSK's advantage is that it is lightweight and runs comfortably on modest hardware. Most current tools have moved to Whisper or whisper.cpp; nerd-dictation is the main VOSK holdout.
What is the best dictation software for Ubuntu?
Vibe Typer. Ubuntu has defaulted to Wayland since 22.04, and Vibe Typer runs natively on it, GNOME included, from one AppImage on Ubuntu 22.04 and later. Canonical's own speech to text preview, Myna, covers GNOME on Wayland only. Details and a comparison in the Ubuntu guide.
Try Vibe Typer free
Voice typing that works in every app on Linux, Windows, and macOS. Free to download: 2,000 words a month, no card.
Download free
