In short
With no language selected, online recognition tries to identify what you speak. This is useful when an English message follows a German one. If short recordings are often misidentified, set your language under General → Languages.
Choose one or several languages
Use Add language to enter the languages you dictate in. Remove one with its cross; an empty list restores automatic detection. A single language hint is especially useful for short sentences.
Several entries tell recognition which languages to expect. They do not guarantee that other languages are excluded. Results for mixed-language sentences also depend on the model.
Check an offline model’s languages
| Model | Languages and selection |
|---|---|
| Small | English, German, Spanish and French; no automatic language selection. |
| Medium | 25 European languages with automatic detection. |
| Large | Around 100 languages with automatic detection. |
Small uses the first supported language in your selection, falling back to German without a suitable preference. Check the language notice beside the model before recording. See offline setup.
Recognition, translation and interface language
The menu language is set under Appearance. It does not determine your dictation language. Default text actions preserve the transcript’s language; translation needs an instruction that asks for it.
If individual foreign names are wrong, a dictionary entry may help more than changing the language of the whole dictation.
Related questions
My language is not in the selection menu. What can I do?
Try automatic detection with an empty language list. Recognition depends on the provider and model. Offline models are limited to their own supported languages.