India · Technology
How BHASHINI turns language data into speech and translation services
India’s BHASHINI programme reported more than 360 language models and 22 specialised services in August 2026. Its published figures describe different parts of the system, from sentence pairs used in language work to tools that turn speech into text.
Sources checked
- 01Parallel sentence pairs
246 million pairs · Text aligned across languages for translation work Source 1
- 02Monolingual text
3.7 million entries · Text resources in a single language Source 1
- 03AI language model repository
More than 360 models · Models supporting specialised language tasks Source 1
- 04Language services
More than 22 specialised services · Functions include recognition, translation and speech generation Source 1
- 05Service coverage, March 2026
36 text languages and 23 voice languages · Coverage varies by modality Source 2
- 06Shrutlekh
22 Indian languages · Live transcription and translation tool Source 3
The flow shows where data, models and applications fit. Individual applications can use different paths.
The data include text in one or more languages
The August parliamentary reply reported 246 million parallel sentence pairs and 3.7 million monolingual text entries in the corpus. A parallel pair aligns two renderings of a sentence across languages. Monolingual entries provide text in one language. These are counts of language resources, rather than people served. Source 1
Models support different types of tasks
The same reply listed more than 360 models and more than 22 specialised services. Speech recognition converts audio to text. Machine translation converts text between languages, while text-to-speech produces audio. Transliteration changes the writing system and optical character recognition converts images of text into machine-readable text. Source 1 Source 4
Voice coverage and text coverage differ
The March platform update reported 36 text languages and 23 voice languages. The March parliamentary reply separately reported that the Shrutlekh real-time transcription and translation tool supports 22 Indian languages. These counts describe the platform and one application at different levels. Source 2 Source 3
The tools reach users through other services
BHASHINI was launched in 2022 under the National Language Translation Mission. The ministry cites Shrutlekh for multilingual event transcription and Maharashtra’s MahaVISTAAR-AI for agricultural advice in a multilingual conversational interface. The platform’s shared capabilities let a host application choose the language tasks it needs. Source 4 Source 3