BodhanAI from IIT Madras launches Indic-Transcribe, works with 26 local languages and English

BodhanAI, an initiative incubated from IIT Madras, has announced Indic-Transcribe – a new transcription model that works with 26 local languages and English. BodhanAI says that it is more accurate that Sarvam's Saaras V3 model as per benchmark scores.BodhanAI has announced Indic-Transcribe, a new model for speech recognition. (Photo: BodhanAI)BodhanAI, an initiative from IIT Madras, has just launched Indic-Transcribe, a speech recognition model built with AI4Bharat that supports 26 Indian languages along with English.The non-profit initiative says that Indic-Transcribe is designed for India. The model is tailored to work across India’s languages, scripts and accents. The model is trained on 1.2 billion parameters.Announcing the launch on X, Professor Mitesh Khapra wrote, “India is a voice-first nation: countless languages, scripts, and accents. Introducing Indic-Transcribe: built to understand India as she actually speaks.” The tool will be available on the BodhanAI website from September 5. A screenshot of Mitesh Khapra's X post. Indic-Transcribe comes in two models – Core and Flex. BodhanAI says that Core is meant for best accuracy, while Flex is aimed at wider coverage, including mixed scripts and Romanised transcriptions. Benchmark numbers shared by BodhanAI said Indic-Transcribe Core recorded a word error rate of 8.9, lower than Sarvam Saaras V3’s 10.1. Indic-Transcribe Flex recorded 11.1, compared with Google Gemini 3 Pro’s 18.9. Benchmark scores shared by BodhanAI. How does BodhanAI’s Indic-Transcribe works?Do note that speech recognition has become a point of focus for AI companies, particularly for Indian use cases, largely due to the diversity in languages and accents in the country. BodhanAI says that Indic-Transcribe is built exactly for this purpose. In the launch video, BodhanAI stated that the supported languages for Indic-Transcribe includes Bengali, Odia, Bhojpuri, Punjabi, Nepali, Haryanvi, Maithili and Hindi, among others.It is designed to recognise different accents and transcribe speech accordingly. The model can also transcribe Sanskrit shlokas, and handle mixed scripts such as cricket commentary with words from two languages or railway announcements. BodhanAI says that you can even transcribe songs with Indic-Transcribe.The launch details also said the system covers remote languages such as Bodo from Assam and Santali from Jharkhand. According to BodhanAI, pilots are now live in 12 languages with 12,000 hours of speech from students of grades 1 to 10. The non-profit states that the parental consent was taken for the pilot programme.In this case you are unaware, BodhanAI is the national Centre of Excellence for Artificial Intelligence in Education and operates under the IIT Madras Bodhan AI Foundation. Backed by the Ministry of Education, it is expected to play a central role as AI tools are integrated across the education system.- EndsPublished By: Armaan AgarwalPublished On: Aug 18, 2026 10:41 IST