Voice Pro recognises and transcribes speech in 30 transcription languages plus 38 translation languages, and the entire app interface — menus, settings, errors, licence dialog — is translated into each one. Same for the website: every page you are reading exists in 35 other languages at /xx/ URLs.
The list
Arabic, Bulgarian, Cantonese, Chinese (Simplified), Croatian, Czech, Danish, Dutch, English, Filipino, Finnish, French, German, Greek, Hindi, Hungarian, Indonesian, Italian, Japanese, Korean, Macedonian, Malay, Persian, Polish, Portuguese, Romanian, Russian, Serbian, Slovak, Slovenian, Spanish, Swedish, Thai, Turkish, Ukrainian, Vietnamese.
How language detection works
Two separate systems:
- Website: Geo-IP lookup on first visit. If you are browsing from Poland, you land on
/pl/. There is a language switcher in the nav to pick another one manually; your choice is remembered. - App (speech recognition): Set your primary language in Settings. The ASR model uses this as a strong hint — it can still recognise occasional English words inside a Polish sentence, but it will not suddenly switch to Czech because you said "hello".
What "fully translated" actually means
Every UI string — every button, menu item, error message, tooltip, settings label, log line the user sees — is stored as a key in a translation catalog. Each language has a file with 275–320 translated strings depending on which features it touches. When the app starts, it loads the catalog for your chosen language and every label in the interface comes from that file.
For the website we do the same thing: one English template per page, a JSON file per language, and a build script that stitches them together. We have 288 generated pages right now (8 pages × 30 transcription languages plus 38 translation languages) and a sitemap that points Google at each translated version with correct hreflang tags.
Translation quality
Initial translations were done by a large model with human review for the top 10 languages (German, Spanish, French, Italian, Polish, Portuguese, Russian, Chinese, Japanese, Korean). For the other 26 we relied on model output alone — which is good but not perfect.
If you see a translation that reads awkwardly, is wrong in context, or sounds like a robot wrote it: tell us. Send the string, the language, and a better suggestion to [email protected]. We are also accepting translation fixes as a valid Champions Program contribution.
Missing language? If there is demand for it and a Qwen ASR model covers it, we will add it.