BodhanAI, an initiative incubated at IIT Madras, has announced Indic-Transcribe, a new AI-powered speech recognition model developed with AI4Bharat.
The model has been designed specifically for India's diverse linguistic landscape, including regional languages, different accents, multiple scripts and mixed-language speech.
According to BodhanAI, Indic-Transcribe supports 26 Indian languages along with English and is trained with 1.2 billion parameters.
What is BodhanAI's Indic-Transcribe?
Indic-Transcribe is a speech recognition and transcription model built to convert spoken language into text.
BodhanAI says the model has been developed around Indian use cases, where speech can involve regional accents, code-switching, different scripts and languages that are often poorly represented in mainstream speech-recognition systems.
Professor Mitesh Khapra announced the model on X, describing India as a "voice-first nation" with numerous languages, scripts and accents.
The model is expected to become available through the BodhanAI website from September 5.
Indic-Transcribe Core vs Flex
BodhanAI has released Indic-Transcribe in two versions: Core and Flex.
Indic-Transcribe Core
The Core model is designed to prioritise transcription accuracy.
BodhanAI's benchmark data shows a word error rate of 8.9, compared with 10.1 for Sarvam's Saaras V3.
Indic-Transcribe Flex
The Flex model is designed for broader language and transcription coverage, including mixed scripts and Romanised text.
BodhanAI reported a word error rate of 11.1 for Flex, compared with 18.9 for Google Gemini 3 Pro in the benchmark shared by the organisation.
However, these figures are benchmarks published by BodhanAI, and comparisons can depend on the datasets, evaluation methodology and test conditions used.
Which Indian languages does Indic-Transcribe support?
According to BodhanAI, Indic-Transcribe supports 26 Indian languages and English.
The languages highlighted in the launch include:
-
Bengali
-
Odia
-
Bhojpuri
-
Punjabi
-
Nepali
-
Haryanvi
-
Maithili
-
Hindi
-
Sanskrit
-
Bodo
-
Santali
The model is also designed to handle different regional accents and variations in spoken language.
This could be particularly relevant for applications where users speak in regional languages rather than English or Hindi.
From Sanskrit shlokas to cricket commentary
BodhanAI says Indic-Transcribe is designed to handle more than conventional spoken conversations.
The model can reportedly transcribe Sanskrit shlokas, multilingual speech and mixed-script content.
Examples highlighted by the organisation include cricket commentary containing words from multiple languages and railway announcements.
BodhanAI also claims that the system can transcribe songs.
The broader goal is to make speech recognition more useful across India's everyday linguistic environments rather than restricting it to standardised speech.
Indic-Transcribe covers remote Indian languages
BodhanAI has also highlighted support for languages that have comparatively fewer digital resources.
The organisation specifically mentioned Bodo, spoken in Assam, and Santali, spoken across parts of eastern India.
According to BodhanAI, pilots are currently running across 12 languages using around 12,000 hours of speech from students in Classes 1 to 10.
The organisation said parental consent was obtained for the pilot programme.
Why Indian-language speech AI matters
India's language diversity presents a significant challenge for speech-recognition technology.
A system trained primarily on English or widely used Indian languages may struggle with regional accents, code-switching and languages with fewer training datasets.
Indic-Transcribe is aimed at addressing that gap by focusing specifically on India's linguistic diversity.
This could have applications across education, government services, customer support, accessibility, media, public transport and digital content creation.
What is BodhanAI?
BodhanAI is associated with the IIT Madras Bodhan AI Foundation and serves as the national Centre of Excellence for Artificial Intelligence in Education.
The initiative is backed by the Ministry of Education and focuses on developing and deploying AI technologies for India's education ecosystem.
Indic-Transcribe is therefore part of a broader push to develop AI systems that are better adapted to India's languages and educational requirements.
