IIT Madras Bodhan AI has launched Indic-Transcribe, a new AI speech recognition model made for India’s many languages. It was developed in collaboration with AI4Bharat.The model supports 26 Indian languages and English and has 1.2 billion parameters. It is designed to understand different Indian accents and languages more accurately. Indic-Transcribe comes in two versions: Core and Flex. Core focuses on better accuracy, while Flex supports more languages, mixed scripts and Romanised text. The model can be used for different purposes, including Sanskrit shlokas, cricket commentary and railway announcements. In tests, Indic-Transcribe Core recorded a word error rate of 8.9, compared with 10.1 for Sarvam Saaras V3. A lower score indicates better accuracy. BodhanAI plans to make Indic-Transcribe available on its website from September 5, 2026.
IIT Madras Bodhan AI Indic-Transcribe: AI Model Supports 26 Indian Languages
