BodhanAI Indic-Transcribe

Overview

BodhanAI, incubated at IIT Madras, launched Indic-Transcribe on August 18 as a speech recognition model supporting 26 Indian languages plus English.[14] Built with AI4Bharat, it reports a word error rate of 8.9, outperforming Sarvam Saaras V3’s 10.1 on shared benchmarks, and is aimed at call centers, media houses, and government projects needing accurate Indic ASR.[14] The tool exposes this model in a usable product for transcription workflows across India’s major languages.