About Vartana Voice

Millions of people move to Bengaluru and never learn to speak Kannada. Not from indifference — because every existing option teaches you to recognise words rather than say them, and because saying a new language badly in front of a stranger is uncomfortable enough that most people stop trying.

How the scoring actually works

When you say a word, two things judge it and the results are combined:

• 60% — phonetic comparison. Your transcript and the target word are both converted to a common romanisation and compared character by character. This part is deterministic: the same attempt always scores the same, so the pass mark cannot drift.
• 40% — a language model. Judges grammar, word choice and word order, and writes the sentence explaining what you said versus what you should have said.

You need 70% to pass. We deliberately do not fold together sounds that actually distinguish words in Kannada — if you say istu when the word is estu, you will be told, because that difference is the entire point of learning to speak rather than to gesture.

About the Kannada itself

Lesson content is reviewed by native Kannada speakers before it reaches learners. This matters more than it might sound: an AI can generate plausible-looking Kannada that no one in Bengaluru actually says, and teaching someone a phrase that marks them as a foreigner is worse than teaching them nothing.

If you are a Kannada speaker and something we teach sounds wrong to you, please tell us at [email protected]. That feedback is more valuable to us than almost anything else.

Who is behind it

Vartana Voice is built and run from Bengaluru, India. It is independent and self-funded. It is not affiliated with any government body, university, or language authority, and we do not issue certificates or qualifications.

What stage this is at

A beta, with a small group of learners in Bengaluru. Hindi → Kannada is live. Tamil and Telugu are planned but not built. If something is broken, telling us is the fastest route to it being fixed.