AI voice generation in 34 languages

These twelve options — Hindi, Hinglish and ten regional languages — share a generation path built for Indian speech. Two practical consequences: reference recordings are capped at 12 seconds, and English words embedded in the text are converted to the local script before generation instead of being silently dropped.

English reference recordings are capped at 12 seconds, the same cap as the Indian-language models. English is the only language available on every tier including free. Read the English voice generation page for what that means in practice, including accent handling.

These twenty-one languages are generated with a multilingual model that supports cross-lingual cloning, and take a 12-second reference clip just like Indian languages. Each has its own page covering what is different about generating it — how the script handles stress, which distinctions your typing has to get right, and whether the accent is something you choose or something your reference clip decides.

The Mayantar interface itself is translated into 11 languages — English, Hindi, Odia, Assamese, French, Spanish, German, Portuguese, Russian, Arabic (right-to-left) and Japanese. That is deliberately separate from the 34 generation languages above: you can run the app in English and generate Tamil, or run it in Arabic and generate Spanish. Nothing about the interface language constrains what you can generate.