2h ago
Linux: option to never translate — transcribe in the spoken language (per-dictation language detection)
I tested Superwhisper for Linux 1.1.3 (Omarchy, cloud mode, free tier) and dictate in both German and English.
Problem:
The output is always in the one selected language. When I speak another language, the speech is translated instead of transcribed:
- Language "English", I say "Das ist ein Test" → it types "This is a test."
- Language "English", I say "Eins, two, three" → "One, two, three."
- "Automatic" does not detect the spoken language; it uses the desktop locale (en_US here), so it behaves exactly like "English".
The history metadata has `translationEnabled: false` for these dictations, yet the text is translated, and there is no setting to prevent it.
Request:
1. A setting "Never translate / keep the spoken language", so a dictation is always written in the language that was spoken.
2. Make "Automatic" detect the language per dictation (as Whisper does), instead of following the desktop locale — optionally limited to a list of languages I choose (e.g. German + English).
Why it matters: many people switch between their native language and English several times a day. Today they have to change the language by hand before each dictation, or they get a translation they did not ask for.
Pending