This page describes Dictum 0.5.

Languages¶

Whisper, the speech model Dictum uses, knows about 99 languages. Each profile has its own language setting, in one of three modes:

One language. Whisper always writes in it. This is the most reliable choice if you dictate in one language.
Several languages. For each dictation, Whisper picks one of them. The first one is the main language: it wins whenever Whisper is unsure.
Auto. Whisper detects any language on its own. It is slower and less accurate for mixed speech.

This page shows where to set the languages, how the main language decides, and what changed in 0.4.

Choose your languages in the setup wizard¶

The setup wizard opens on the first start. You can run it again later: Settings → General → Setup wizard → Run setup again.

Step 1, Which languages do you dictate in?, lists every language Whisper knows. On a first setup, your system's language is already ticked, and English too.
Tick each language you dictate in. Type in Find a language… to find one by its own name, its English name or its code, such as uk.
One of the ticked languages wears the Main badge. To make another one main, click Make main on its row.
If you only ever dictate in one language, untick the others.
To let Whisper detect any language instead, tick Or let Whisper detect any language (slower, less accurate for mixed speech).
Click Next.

The wizard sets the languages of your first profile. Other profiles keep their own: change them in Settings → Profiles.

Change the languages of a profile¶

Click the Dictum icon in the tray, or choose Settings in its menu.
Open Profiles and select the profile on the left.
Under Languages, pick the mode: One, Several or Auto.
- One: choose the language from the list.
- Several: each language is a chip. The first chip is marked main. Click Make main on another chip to put it first. Add a language with Add a language, and remove one with the × on its chip. The last chip cannot be removed.
- Auto: "Whisper detects the language of each dictation."

There is no Save button. Settings saves a change half a second after you make it, and the running Dictum picks it up for the next dictation.

In config.json, the setting is the profile's language. With several languages, the main one comes first:

"language": "uk, en"

How the main language decides¶

With several languages, Whisper first listens to the recording and says which of your languages it hears, and how sure it is. Dictum then picks the language to write in:

If Whisper hears the main language, Dictum uses it.
If Whisper hears another language from your list and is at least 50% sure of it, Dictum uses that language.
In every other case, Dictum uses the main language.

For example, with uk, en:

You dictate in Ukrainian: the text comes out in Ukrainian.
You dictate a whole sentence in English: Whisper is usually sure of it, and the text comes out in English.
Whisper hears English but is only 6% sure, which happens with Ukrainian speech: the text comes out in Ukrainian.

So make the main language the one you dictate in most.

Each choice is written to Dictum's log, dictum.log in the logs folder (%LOCALAPPDATA%\Dictum\logs on Windows, ~/.local/share/dictum/logs on Linux). A line where the main language won looks like this:

2026-10-21 10:15:02.481 +02:00 DEBUG Whisper 'local' picked 'uk' from uk, en (detected 'en', p=0.06; the first language has priority, and another one needs p >= 0.50) in 38 ms

Closely related languages¶

Whisper often mixes up languages that sound alike: Ukrainian, Belarusian and Russian; Polish, Czech and Slovak; Spanish, Portuguese and Galician; Norwegian (both forms), Danish and Swedish; Croatian, Serbian and Bosnian; Indonesian and Malay; Hindi and Urdu.

If you put two of them in one profile, the wizard and the Profiles page warn you. The text may come out in the other alphabet or spelling. Keep the one you use most in this profile, and put the other in a profile of its own, with its own hotkey: Profiles shows how.

Why words came out "translated" before 0.4¶

Before 0.4, Dictum took whichever of the profile's languages Whisper rated highest, however unsure Whisper was. Ukrainian speech often makes Whisper think of Russian first. With uk, en and no Russian in the list, English could end up on top with a probability as low as 6%. Whisper then wrote your Ukrainian speech in English: the text came out translated.

Since 0.4, the main language wins unless Whisper is at least 50% sure of another one, as described above. If your text still comes out in the wrong language, see Troubleshooting.

A remote speech server gets only the main language¶

A profile can use a remote server for recognition instead of Whisper on your computer: you add one with + Remote server on the Speech providers page of Settings, and choose it as the profile's Speech provider.

An OpenAI-compatible server takes one language per request, not a list. So with several languages, Dictum sends only the main one, and the server writes everything in it. The Profiles page shows a warning under Languages, in the config validator's words, for example:

'remote' takes a single language, so with a list the server gets the first language, 'uk'

With Auto, Dictum sends no language, and the server detects it.

If you dictate in two languages through a remote server, make two profiles, each with one language.

Try it from the command line¶

dictum transcribe runs a WAV file through recognition, so you can compare settings on the same recording:

dictum transcribe memo.wav --lang "uk, en" --raw
dictum transcribe memo.wav --lang uk --raw

--lang takes one code, a list such as "uk, en", or auto. --raw skips the profile's processors, so you see what Whisper wrote. dictum profiles show dictate prints a profile's language:.

Note

On Windows, the MSI does not put dictum on PATH. In PowerShell, run it by its full path, for example & "$env:LOCALAPPDATA\Programs\Dictum\dictum.exe" transcribe memo.wav --raw. In the AppImage, put dictum after the file name: ./dictum-<version>-linux-x64.AppImage dictum transcribe memo.wav --raw.