For Linux

Linux dictation.
That actually exists.

Wispr Flow doesn't run on Linux. Willow doesn't. Voibe, VoiceInk, Superwhisper, Monologue: none of them. Voca does.

Most "best dictation app of 2026" lists are really "best macOS dictation app of 2026" lists with the qualifier missing. The mainstream commercial options ship for Mac, sometimes Windows, almost never Linux. If you've been using nerd-dictation, Numen, Talon, or one of the Whisper-based local scripts because that was the only Linux option available, this page is for you.

Voca ships a real Linux build. Same hotkey-and-speak workflow as the Mac and Windows versions. System-wide, works in any text field. $6 a month.

Try it in your browser Download for Linux

No install for the demo, no card required.

01 · The gap

Why Linux dictation
has been so bad.

Three reasons, none of them good.

The result is that Linux developers, who arguably benefit most from voice dictation (heavy AI prompting workflows, lots of CLI work, a culture comfortable with hotkeys), have had the worst tooling. Voca is the same hosted Whisper-class accuracy you'd get on Mac, with a Linux build that works.

01 · VENDORS

Most commercial vendors don't bother.

Linux is a smaller market, the GUI fragmentation between GNOME, KDE, X11, and Wayland makes building a polished desktop app harder, and the engineering investment doesn't always pencil. So they ship for Mac and call it done.

02 · LOCAL WHISPER

The local Whisper option is real but rough.

Talon, Numen, nerd-dictation, whisper.cpp wrappers all work, but they require setup, and accuracy without a fine-tuned model is uneven on technical vocabulary.

03 · NO WORKAROUND

The "use macOS dictation" workaround obviously doesn't apply.

If you're on Linux because you chose Linux, you're not switching to a Mac to dictate.

02 · What it runs on

Distros
and desktops.

Voca's Linux build runs on the major distros: Ubuntu, Pop_OS, Debian, Fedora, Arch, openSUSE, and most of their derivatives. It works under both X11 and Wayland.

If you're on a less common distro, the AppImage is the safest bet for compatibility.

Download for Linux

03 · Where it works

Apps
and surfaces.

System-wide. Wherever there's a text field, there's a place Voca can put text. The most common surfaces:

Anywhere your cursor sits, transcribed text arrives.

01 · IDEs

IDEs.

VS Code, Cursor, JetBrains family (IntelliJ, PyCharm, GoLand, etc.), Sublime, Zed, Helix.

02 · TERMINALS

Terminals.

GNOME Terminal, Konsole, Alacritty, Kitty, Wezterm, Foot, Ghostty.

03 · BROWSERS

Browsers.

Firefox, Chromium, Brave, LibreWolf, Vivaldi.

04 · COMMS

Communication.

Slack, Element, Discord, Zulip, Teams (web).

05 · NOTES & DOCS

Notes and docs.

Obsidian, Logseq, Joplin, Standard Notes, LibreOffice, OnlyOffice.

06 · EMAIL

Email.

Thunderbird, web Gmail, web Fastmail, web Proton.

04 · Privacy

A posture
that actually matches
the audience.

Linux users tend to be more privacy-aware than the median desktop user. Voca's design is the simpler version: the microphone is on while you hold the hotkey, the audio goes to our server, transcription happens, audio is deleted, text comes back. Nothing else.

  • Audio deleted as soon as the transcript returns. No backups, no logs of what you said.
  • No model training on your voice or transcripts.
  • Transcript history is opt-in.
  • Metadata for billing only — duration, timestamp, model. Nothing more.
01 · Client
Linux desktop
Captures audio when you hold the hotkey. X11 or Wayland.
02 · Server
Voca API
Transcribes the audio. Deleted as soon as text returns.
03 · Return
Transcript
Text appears at the cursor in any text field on your machine.
Audio retained0 sec · always
Three steps

Hold. Speak.
Done.

01 · CAPTURE

Press the hotkey on Linux.

In your terminal. In your IDE. In your browser. In Slack. In any text field.

CtrlShift;
02 · SPEAK

Talk like a human.

Filler words removed. Formatting inferred. 90+ languages supported.

"Add a flag to skip the migration check when running locally…"
03 · RETURN

Text lands at your cursor.

0.4 seconds, median.

audio → transcript → ✓ pasted
Linux options compared

What's the alternative
on Linux, really?

Feature
Voca
Talon / Numen / nerd-dictation
macOS-only commercial tools
Runs on Linux
Setup effort
Install + hotkey
Variable, can be significant
N/A
Accuracy on technical vocabulary
High (hosted Whisper class)
Variable, model-dependent
N/A
Median latency
0.4s
Variable
N/A
Browser fallback
N/A
Cost
$6/month
Free / open source
N/A

The local-Whisper options are great if you have the patience to maintain them. Voca is for the days you don't. Also worth a look: Voca vs Wispr Flow.

Try Voca on Linux.

The browser version runs in any modern browser without an install. Useful for testing accuracy before pulling the desktop binary down.

Try it in your browser
FAQ

Linux users,
the actual questions.

Does Voca support Wayland?

Yes. Voca runs under both X11 and Wayland on supported distros.

Which Linux distros are supported?

The most common ones (Ubuntu, Pop_OS, Debian, Fedora, Arch, openSUSE) and most of their derivatives. The AppImage build runs on most distros that aren't explicitly on the list.

Is the Linux build a first-class citizen, or a port?

First-class. The Linux build is built and shipped at the same cadence as the Mac and Windows builds, with the same feature set.

Does Voca work in my terminal?

Yes. Voca works in any text field, including terminal emulators (GNOME Terminal, Konsole, Alacritty, Kitty, Wezterm, Foot, Ghostty). Useful for typing long shell commands, commit messages, and prompts to CLI agents like Claude Code.

Does it work with VS Code, Cursor, and JetBrains IDEs on Linux?

Yes. All of them have native Linux builds, and Voca operates at the OS level, so transcribed text appears wherever the cursor is in any of them. See Voca for Cursor.

What's the cheapest way to try it?

The browser demo runs without an install or a card. After that, the desktop app gives 30 free minutes of transcription per month on the free tier.

Are my recordings stored?

No. Audio is deleted on transcription. We don't train on it, log it, or back it up. Metadata for billing (duration, timestamp, model) is the only thing kept.

Linux dictation,
without the duct tape.

$6/month flat or $0.10/hour pay-as-you-go. Same hotkey on Mac, Windows, and Linux. Pairs with Voca for Cursor and Voca for Claude Code. See team pricing for shared seats.

Try it in your browser Download for Linux