Supported languages
Which languages can be captured live, which need an upload, which we have measured, and which aren't supported.
Updated October 30, 2026
Transcription quality differs by language, and we only call a language tested once we've measured it ourselves. The tiers below say exactly what each language has behind it. Checked against our speech provider's published lists on 2026-10-30.
| Tier | Languages | What to expect |
|---|---|---|
| Measured by us | English, Spanish | In our published evaluation set. Spanish is covered by one session so far, so review Spanish drafts closely. |
| Live capture, provider-supported | Afrikaans, Arabic, Cantonese, Catalan, Danish, Dutch, Estonian, Finnish, French, Galician, German, Hebrew, Hindi, Italian, Japanese, Korean, Mandarin, Marathi, Norwegian, Norwegian Nynorsk, Persian, Portuguese, Romanian, Russian, Swedish, Turkish, Urdu, Vietnamese, Xhosa, Zulu | Supported by our speech provider's real-time model. Not yet measured by us: treat the draft as a starting point and check medication names and clinical terms. |
| Uploaded audio only | 55 more languages, listed below | Supported only by our provider's pre-recorded model, so they need audio upload rather than live capture. Many have low published accuracy. Not measured by us. |
| Not supported | Haitian Creole, Hmong | Our speech provider doesn't list these today. Use an interpreter and capture the interpreted language, or write a typed recap. |
Upload-only languages
Amharic, Armenian, Assamese, Azerbaijani, Belarusian, Bengali, Bosnian, Bulgarian, Burmese, Croatian, Czech, Georgian, Greek, Gujarati, Hausa, Hungarian, Icelandic, Indonesian, Javanese, Kannada, Kazakh, Khmer, Lao, Latvian, Lingala, Lithuanian, Luxembourgish, Macedonian, Malay, Malayalam, Maltese, Maori, Mongolian, Nepali, Occitan, Pashto, Polish, Punjabi, Serbian, Shona, Sindhi, Slovak, Slovenian, Somali, Swahili, Swiss German, Tagalog, Tajik, Tamil, Telugu, Thai, Ukrainian, Uzbek, Welsh and Yoruba.
Mixed-language sessions
The real-time model detects and follows a switch between supported languages mid-session; we have measured this only for English and Spanish. Quotes in the note keep the original language with an English gloss when the note is written in English.
Interpreted sessions
Mark the interpreter as a participant. Margent transcribes both languages and attributes interpreted statements to the client, with the interpreter noted.
Was this article helpful?