BodhanAI and AI4Bharat launch Indic-Transcribe recognizer for 26 Indian languages
IIT Madras's BodhanAI and AI4Bharat just dropped Indic-Transcribe, a speech recognition tool that understands 26 Indian languages plus English.
It's built to handle different accents, scripts, and even mixed-language conversations, making it super relevant for India's diverse way of speaking.
You can try it out on the BodhanAI website from September 5.
Core 8.9% WER, flex supports romanized
There are two models: Core (for high accuracy) and Flex (for mixed scripts).
Core nails an 8.9% word error rate, lower than Sarvam Saaras V3's 10.1, while Flex supports Romanized text with solid results too.
The tool can pick up everything from Hindi and Bengali to Sanskrit shlokas and songs.
School trials cover 12 languages
Indic-Transcribe is already being tested with school students across 12 languages using real speech data (with parental consent).
As professor Mitesh Khapra puts it, this is about helping India be understood as it "actually speaks," a step forward for inclusive tech in classrooms and beyond.