Speech, translation and interface languages
These are three different lists. Speech is recognized and translated in 44 languages; the program interface is available in 20 languages.
Main speech and translation languages
No more than 10% of words are misrecognized, on both a graphics card and a processor.
- English
- Bulgarian
- Vietnamese
- Indonesian
- Spanish
- Italian
- Catalan
- Chinese
- Korean
- Malay
- German
- Dutch
- Norwegian
- Polish
- Portuguese
- Romanian
- Russian
- Slovak
- Turkish
- Ukrainian
- Finnish
- French
- Croatian
- Czech
- Swedish
- Japanese
Additional languages
Lower quality: errors in 10–25% of words, medium-quality translation. In the program's lists they come after the main languages and are marked “lower quality”.
- Azerbaijani
- Arabic
- Bosnian
- Hungarian
- Galician
- Greek
- Danish
- Icelandic
- Kazakh
- Latvian
- Lithuanian
- Macedonian
- Persian
- Serbian
- Slovenian
- Thai
- Urdu
- Estonian
Interface languages of the program and the installer
- English
- Arabic
- Bengali
- Vietnamese
- Indonesian
- Spanish
- Italian
- Chinese
- Korean
- German
- Persian
- Polish
- Portuguese
- Russian
- Turkish
- Ukrainian
- Urdu
- French
- Hindi
- Japanese
How we evaluated
Measured with FLEURS recordings on October 25, 2026. For the main languages, the word error rate is 10% or lower on both a graphics card and a processor. For the additional languages it is 10–25%, and translation is of medium quality. For Chinese, Japanese and Thai, errors were counted in characters.
- Hindi and Bengali are available only in the interface: there is no speech recognition or translation for them.
- Auto-detection may confuse Bosnian, Serbian, Macedonian and Galician with neighboring languages, so it is better to select these languages explicitly.
- Filler word lists, voice punctuation and replacement dictionaries are preset for Russian and English.