Speech Note for Linux in 2026: Offline Dictation Test

A practical setup test for local Linux speech-to-text, including the Wayland steps most reviews skip.

Speech Note is getting fresh attention after ZDNET recommended it this week as an offline speech-to-text tool for Linux. It sounds simple: install one Flatpak, download a model, and dictate without sending audio to the cloud.

The more useful answer is a bit more specific. Speech Note works very well for private, local transcription. It can also insert text into other Linux apps, but that depends on your desktop session. On Wayland, inserting text into the active window requires ydotool and the correct Flatpak socket permission. Global shortcuts also require support from your desktop portal.

The question isn’t whether Speech Note can recognize speech. It can. The question is whether it can send that text to the apps you use without making you copy and paste it.

What Speech Note does well

Speech Note is free, open source software for Linux desktops and Sailfish OS. It supports speech-to-text, text-to-speech, and machine translation. According to the project, it processes text and voice locally. After you download the required model files, it doesn't send data to the internet.

It gives you more model choices than most desktop dictation apps. You can use Vosk, whisper.cpp, Faster Whisper, Coqui STT, or april-asr for speech recognition. That flexibility helps, but two people can install the same app and see very different speeds and accuracy.

ZDNET’s Jack Wallen used the English Faster Whisper Medium model and praised its accuracy. His workflow was simple and focused on taking notes: click Listen, speak, click Stop, and wait for the transcript. It works well for brainstorming, draft monologues, and private notes.

Hackaday published a more cautious report that same day. The author found Speech Note less seamless on his system and preferred another local tool. That difference matters. Linux speech recognition can vary widely based on your CPU architecture, graphics support, desktop environment, model size, and how you want the text delivered.

Install Speech Note with Flatpak

The usual way to install it is through Flathub:

flatpak install net.mkiol.SpeechNote

Speech Note doesn’t include speech models with the app. After installing it, open the model browser and download at least one speech-to-text model for your language.

Start small unless you know your hardware can handle a larger model. A small model gives you a quick latency baseline. Then test a medium model with the same 60-word recording. Keep the larger model only if its accuracy gains save more time than the extra processing delay costs.

Storage may be the first surprise. The project's full Flatpak base package is much larger than its Tiny package, and optional GPU add-ons can take up a lot of disk space. Check the current package sizes in the official README before installing an accelerator just because your machine has an AMD or NVIDIA GPU.

The active-window feature changes the comparison

Several reviews present Speech Note mainly as a private notepad. The current project documentation says it can do more. Speech Note can copy decoded text to the clipboard or insert it into the active desktop window.

Use this command to listen to the active window:

flatpak run net.mkiol.SpeechNote --action start-listening-active-window

On X11, active-window insertion should work without extra setup. On Wayland, you need the ydotool daemon and must give the Flatpak access to its socket. That difference may decide whether the app stays a nice demo or becomes reliable enough for daily use.

Speech Note also supports global keyboard shortcuts when it runs in the background. On Wayland, this feature uses the GlobalShortcuts interface from the XDG Desktop Portal. The project currently lists recent KDE Plasma and GNOME desktops as supported environments.

Test the actual destination before you decide the setup works. If a transcript appears in Speech Note, you know the model decoded your audio. You still don’t know whether that text will reach Firefox, LibreOffice, Slack, an IDE, or a remote desktop field.

Use a real accuracy and latency test

Don’t judge a dictation tool with a clean sentence about the weather. Test it with a sample that resembles your work. Include two names, an email address, a number, one acronym, and three terms that generic models often miss.

Record the same sample five times. Track four things:

Time from stopping speech to seeing the final text
Number of wrong words
Time spent fixing names and specialist terms
Whether the text reaches the intended app without a manual paste

Then run the test again with a second model. A model that makes fewer errors may still take longer if it pauses for several seconds after each recording. A faster model may be worse if you have to fix every paragraph.

Speech Note 4.8 can also use an existing note as context when decoding. This may help it keep terms consistent throughout the note. However, the project warns that context can sometimes reduce quality. Test it with and without context instead of assuming it will always help.

Speech Note is not streaming dictation

ZDNET's test highlights an important detail: the text didn't appear while he was speaking. It showed up only after he clicked Stop and Speech Note processed the recording.

That delay works for notes and longer thoughts. It feels more awkward when you use live text to catch mistakes before finishing a sentence. Recent discussions among local dictation users keep circling back to the same request: streaming feedback and the ability to mix speech with keyboard edits.

Treat final text speed and live feedback as separate features. A tool may finish quickly after you stop speaking but still feel unclear during recording. That may be fine for short messages. For technical writing or accessibility, visual feedback can help you stay in control.

Where Speech Note fits, and where DictaFlow does not

Speech Note is the best choice for Linux users who want local, offline transcription and don't mind configuring desktop integration. DictaFlow doesn't currently offer a Linux app, so it isn't a direct replacement.

If you work across Windows, Mac, iPhone, iPad, or Android, the tradeoff is different. DictaFlow lets you hold a key and dictate directly into the app you’re using. It includes custom vocabulary, cleanup based on the app, and typing-style input for Citrix, RDP, VMware Horizon, and fields that block clipboard pastes. Pro costs $7 per month or $69 per year.

The broader dictation software comparison covers the platform differences. The DictaFlow getting started guide explains how to set up hold-to-talk on supported devices.

Use Speech Note if Linux support and offline processing matter most. If you switch between Linux and another operating system, test the same sample on both. Choose the setup that handles your vocabulary and gets the text into the final field with the least cleanup.