BHASHINI and the National Language Translation Mission

India has 22 officially recognised languages and hundreds of dialects, yet content in local languages online is scarce, and much available language data is not in a form that can train AI models. The government's response was the National Language Translation Mission (NLTM), announced in the Union Budget for 2021-22, and BHASHINI, the platform through which it is implemented.³ BHASHINI was launched in July 2022 at Digital India Week in Gandhinagar.¹ It aims to use AI and language technologies to give Indians easy access to the internet and digital services in their own languages and to increase Indian-language content.³⁶

NITI Aayog's 2018 national AI strategy had already suggested advanced natural language processing infrastructure for Indian languages as an example of a 'moonshot' research project,¹⁰ and MeitY's 2025 governance guidelines recommend locally relevant datasets to support culturally representative models.⁸

BHASHINI is funded and steered by MeitY. An apex committee led by the MeitY Secretary sets goals, and an executive committee oversees foundation operations, research and ecosystem outreach.³ It runs as the Digital India BHASHINI Division (DIBD), an independent business division under the Digital India Corporation, headed by CEO Amitabh Nag, to work with startups and the wider ecosystem.¹⁴⁶

Its main components are:³¹

• ULCA (Unified Language Contribution API), a repository for language data, training datasets, models and tools contributed by government, startups, civil society and companies; components built under the project are meant to be open source.
• Bhasha Daan, a crowdsourcing platform on which people can transcribe audio, record audio of text prompts, translate text or label images.
• Open BHASHINI APIs, through which more than 300 pre-trained AI models for tasks such as translation and speech recognition are offered to ecosystem partners, according to Wikipedia.
• Applications such as a translation tool and, in 2023, a WhatsApp chatbot (initially in twelve languages) under development to answer questions about government schemes, including by voice note.

The mission's white paper stressed attention to low-resource North-East and tribal languages, to be pursued with state governments and local institutions.⁴

Public profile grew quickly after launch. In February 2023 Microsoft CEO Satya Nadella described a BHASHINI demonstration, in which a Hindi-speaking farmer asked about a subsidy and the system recognised his speech and pointed him to government portals, as "the most profound thing that I've seen in a long, long time".³ Microsoft used BHASHINI in its Jugalbandi chatbot for rural users.³⁴ In December 2023 a speech by Prime Minister Narendra Modi was translated in real time for a Tamil audience using BHASHINI, and Finance Minister Nirmala Sitharaman used it during her 2024 Budget speech, according to Wikipedia.¹

Partnerships and integrations reported include: the Reserve Bank Innovation Hub (October 2023, to extend digital financial services in native languages), UiPath (January 2024), Snapdeal (June 2024), the Ministry of Information and Broadcasting, Federal Bank, ONDC's Saarthi reference app for multilingual buyer apps (September 2024) and Rajasthan's Pehchan app for birth and death registration.¹⁶ MeitY's 2024 year-end review reported 50-plus stakeholders on board, including NPCI, RBIH, the Ministry of Rural Development, the Lok Sabha and the Rajya Sabha, with over 5 lakh downloads and 17 languages supported.⁷

On 31 January 2025 Tripura signed a memorandum of understanding with the Digital India BHASHINI Division to integrate it into the CM Helpline, eVidhan, the Kisan Sahayata App and e-District services, the Amar Sarkar app, FIR translation and voice-based data entry.⁴ In June 2024 MeitY also sought research proposals from colleges and universities to improve machine translation accuracy, including domain-specific models for government, education and healthcare.¹

At the India AI Impact Summit on 18 February 2026 the Digital India BHASHINI Division launched VoicERA, an open-source, end-to-end voice AI stack running on BHASHINI's national language infrastructure, with development partners EkStep Foundation, IIIT Bengaluru, AI4Bharat and the Centre for Open Societal Systems. It supports real-time speech processing, conversational AI and multilingual telephony, deployable in the cloud or on premises, with government use cases in agricultural advisories, education support, grievance redress and scheme discovery.²

A policy report released alongside it, 'Building an Open and Responsible Voice Technology Ecosystem', recommended that:²

1. foundational speech datasets be treated as public goods, with representative funding that prioritises low-resource and tribal languages, publicly funded datasets open by default subject to privacy and IP safeguards, clearer copyright and data-protection guidance (including research exemptions and publicly available personal data) and coordinated repositories with tiered access;
2. nationally coordinated evaluation datasets and leaderboards reflect dialects, code-switching and noise, with pooled public and academic compute through shared clusters;
3. dataset hosting be treated as durable public infrastructure, with data cards, model cards and provenance statements for publicly funded projects; and
4. deployment build in value-sharing and attribution for communities and annotators, and counter voice cloning, phishing and deepfakes through technical safeguards, legal recourse and public literacy.

The Digital India BHASHINI Division's CEO described the framework as enabling "secure, scalable multilingual systems", and the National Payments Corporation of India linked voice interfaces to financial inclusion.²

BHASHINI is not a regulatory measure, but it sits at the point where AI, data and language policy meet. MeitY's India AI Governance Guidelines of November 2025 cite "multilingual interfaces like Bhashini" among examples of how digital public infrastructure (DPI) lets AI deliver public goods, and recommend integrating AI with DPI to make solutions scalable and inclusive.⁸ The IndiaAI Mission's AIKosh platform and its sovereign foundation-model programme address the same shortage of Indian-language data (see those entries).

MeitY's Additional Secretary Abhishek Singh said in September 2024 that BHASHINI would need Indian-language datasets for natural language processing projects, that datasets in all languages would be made available so that companies can build AI applications in any sector and language, and that voice-enabled services are "the next frontier of service delivery". He also said that, of about 500,000 government open datasets, "very few" were usable for the AI ecosystem.⁹

The 2026 policy report puts data-protection and copyright questions on speech datasets squarely on the table, including whether a separate lawful basis for AI processing is needed, an area where the governance guidelines also flag open questions under the Digital Personal Data Protection Act, such as exemptions for training on publicly available personal data.²⁸

As of 15 October 2026 BHASHINI is an operating MeitY programme with partnerships across state governments and industry. The sources reviewed did not give current, official counts of languages, models, users or API calls: the 2024 review reported 17 languages and 5 lakh downloads, and the Wikipedia figure of more than 300 models reflects reporting up to September 2024. MediaNama's February 2026 article on VoicERA also gave no figures on models, users or languages, and the budget for the National Language Translation Mission was not identified in the sources reviewed.¹²⁷ The platform's official site is bhashini.gov.in, but its live usage metrics could not be retrieved for this entry.⁵ The programme's openness commitments, its handling of speech data and its stance on voice cloning will depend on how MeitY acts on the February 2026 policy report and on the DPDP Act's phased commencement.