Economic Survey Calls for Voice-First AI System in Indian Languages to Boost Digital Inclusion

Economic Survey Calls for Voice-First AI System in Indian Languages to Boost Digital Inclusion

Why in News? 

The latest Economic Survey 2025–26 highlights the need for a voice-first AI system in Indian languages to enhance digital inclusion, especially for non-English speakers, urging development of accessible, multilingual AI interfaces that can bridge language barriers and bring digital services to every citizen.

Economic Survey 2025–26: Voice AI for Digital Inclusion

• The Economic Survey 2025–26 underscores a transformative vision for voice‑first AI systems in Indian languages as a key element of Digital Public Infrastructure (DPI) to enhance digital inclusion and make technology accessible to historically excluded populations. 
• It states that traditional AI strategies focused on frontier models are less suitable for India’s structural realities, advocating instead for decentralised, application‑driven AI systems that work effectively on low‑cost devices and in native languages to expand reach across socio‑economic groups.
• The survey highlights how voice‑enabled AI solutions—integrated within platforms like Bhashini — can break language barriers in digital engagement, enabling users who do not understand English to interact with services through speech interfaces in local languages. 
• By embedding voice‑first AI systems into DPI such as digital health, agriculture, education, and governance, the Survey sees a pathway to bridge digital divides and ensure that technological benefits are spread beyond urban and English‑speaking citizens. These AI interfaces can interact in Indian languages and deliver information on schemes, markets, and services directly to rural and semi‑urban users.
• The report also stresses that AI in India should be problem‑driven rather than hype‑driven, focusing on real‑world impact, such as assisting farmers with market intelligence, aiding healthcare access, or improving government service delivery. This practical emphasis is part of a broader strategy to build shared infrastructure under initiatives like the National AI Mission, which has been allocated ₹10,300+ crore over five years.
• The Survey recommends that India’s AI ecosystem leverage bottom‑up innovation, with local startups, academic institutions, and community organisations developing context‑specific voice‑based tools that embed vernacular understanding and cultural nuance.

India’s Digital Exclusion Crisis

India stands as the world’s third-largest digital economy in 2025, yet the promise of “Digital India” remains unevenly distributed. Beneath the surface of soaring UPI transactions and 806 million internet users, significant barriers continue to isolate millions based on language, gender, and geography. 

• The Lingual Barrier: Despite hosting 1,600 languages and dialects, the vast majority of digital content remains in English or Hindi, alienating over 44.7% of the population who remain offline. While 22 Scheduled Languages are now gaining ground via AI initiatives like Bhashini, nearly 98% of users now prefer content in Indic languages such as Tamil, Telugu, and Malayalam. This linguistic mismatch creates a psychological barrier, where many perceive digital tools as exclusive to the English-speaking elite. 
• The Literacy Chasm: Basic literacy is a prerequisite for digital literacy, yet a 25% general illiteracy rate hampers progress. A critical gap exists between “usage” and “productive skills”; only 32.2% of youth can create a digital presentation, and a mere 24% of the population is considered financially literate. This lack of functional digital fluency leaves low-income groups vulnerable to cyber fraud. 
• Rural-Urban Disparity: The rural-urban divide remains a primary driver of exclusion. While urban internet penetration has reached 66%, rural connectivity lags at approximately 37%. Rural areas face severe infrastructure bottlenecks, including unreliable electricity and limited bandwidth. Although rural internet subscriptions grew by 200% recently, nearly 40% of the rural population still lacks basic access. 
• Gender Inequality: Indian women face a “double disadvantage” due to intersecting socio-economic and patriarchal norms. Women are 16% less likely to own a mobile phone and 33% less likely to use mobile internet than men. In rural areas, this is often due to social beliefs that treat mobile phones as a risk to a woman’s reputation. Consequently, only 25% of rural women access the internet compared to 49% of rural men. 
• Socio-Economic Stratification: Digital access is increasingly becoming a caste and income privilege. Upper-caste households enjoy internet access rates as high as 94.5% in urban areas, whereas Scheduled Tribe (ST) households face the sharpest disadvantage, with the widest rural-urban gap of 9 percentage points. This hierarchy is compounded by income disparities, leaving marginalized groups with unstable connections. 

India’s Strategic Pivot to Voice-First AI

• Sovereign Strategy: By the IndiaAI Mission’s ₹10,372 crore outlay, the government is building an inclusive infrastructure and is prioritizing sovereign Large Multimodal Models (LMMs) that prioritize voice over text. India’s strategy focuses on multilingual voice automation to handle billions of daily tokens in regional dialects. This “differentiating edge” ensures that even the 490 million informal workers—many of whom face literacy gaps—can interact with digital services as easily as they speak. 
• Digital India Goals: Under the Digital India vision, voice-AI is transforming public service delivery into a localized, real-time experience. Flagship reforms include Kisan e-Mitra, a voice-assistant for farmers, and the integration of AI speech tools into Aadhaar and UMANG, which now offer 2,300 services across 23 Indian languages. These reforms aim to replace complex forms with simple voice commands. 
• Institutional Support and Global Leadership: The government has established four National Centres of Excellence (CoEs) to drive research in healthcare, agriculture, and education. With the AI Impact Summit 2026 on the horizon, India is positioning its “Voice-First” public infrastructure as a global blueprint for the Global South. This is supported by a robust compute pillar that has already onboarded 38,000 GPUs to provide startups with the processing power needed to build voice-native applications. 

Technological Infrastructure for Multilingual AI Systems in India

• Bhashini and BharatGen: India’s AI ecosystem is anchored by the BharatGen initiative, the world’s first government-funded Multimodal Large Language Model (MLLM). Unveiled at the 2025 BharatGen Summit, it integrates text, speech, and image processing for 22 scheduled languages. Complementing this, the Bhashini platform facilitates real-time translation and hosts over 350 AI language models, enabling seamless communication across governance and education.
• Sovereign Cloud and Compute Power: The IndiaAI Mission has established a massive compute pillar to lower barriers for indigenous innovation. The government has onboarded over 38,000 GPUs—including advanced NVIDIA H100 and B200 chips—available to startups and researchers at a subsidized rate of approximately ₹65 per GPU-hour. This sovereign compute power supports data localization, while fueling real-time processing for Automatic Speech Recognition (ASR). 
• AIKosh: AIKosh is a centralized dataset platform that develops large-scale, high-quality Indic datasets. AIKosh currently hosts over 1,000 India-specific datasets and 208 AI models across 20 sectors, including specialized audio for Text-to-Speech (TTS). A landmark feature is the creation of a secure corpus featuring 15,000+ hours of annotated voice data across 22 languages, which allows indigenous AI to understand local dialects and Indian Knowledge Systems (IKS). 
• Speech Recognition and Indigenous Innovation: Indigenous startups and research labs are perfecting Automatic Speech Recognition (ASR) for “Hinglish” and other code-mixed dialects. Through the IndiaAI Innovation Centre, developers are building lightweight AI models. Domestic firms like HFCL and ITI are deploying indigenous IP/MPLS routers for BharatNet. Private startups such as Sarvam AI and Gnani AI are building open-source 120-billion parameter models specifically optimized for India’s linguistic diversity.
• BharatNet: The BharatNet project provides the critical high-speed backhaul for regional AI deployment. With an approved outlay of ₹1.40 lakh crore, BharatNet Phase-3 aims to connect nearly 6.5 lakh villages by 2028. As of January 2026, over 97% of Gram Panchayats have been linked via a 6.83 lakh km optical fiber network. The infrastructure is being modernized using MPLS technology and a ring topology to ensure 100% redundancy and 24×7 monitoring, enabling low-latency AI inference in rural areas. 

Governance and Data Protection Framework for India’s AI Ecosystem

• The DPDP Rules 2025: The Digital Personal Data Protection (DPDP) Rules, 2025, notified in November 2025, fully operationalize India’s landmark privacy law. It enforces the SARAL (Simple, Accessible, Rational, Actionable) compliance approach, requiring organizations to provide privacy notices in 22 regional languages. Data fiduciaries are now legally obligated to report breaches to the Data Protection Board of India within a strict 72-hour window, with penalties reaching ₹250 crore for security failures. 
• Algorithmic Accountability for SDFs: Entities designated as Significant Data Fiduciaries (SDFs)—including major e-commerce and social media platforms—face enhanced obligations. Under Rule 13, these firms must conduct annual independent data audits and Data Protection Impact Assessments (DPIAs).  They are now legally accountable for algorithmic fairness, requiring them to verify that their AI recommendation engines and automated scoring systems do not discriminate against protected groups. 
• Institutional Safety: The newly established AI Safety Institute (AISI) serves as the technical validator for trust and safety standards. It manages a National AI Incident Database to track real-world harms. Supporting this is AIKosh, the national dataset repository, which ensures data localization by housing over 15,000 hours of annotated voice data within sovereign borders.
• The Seven “Sutras” of AI Governance: Released on 5 November 2025, the India AI Governance Guidelines establish seven foundational principles, or Sutras, to guide the AI lifecycle. These include Trust as the Foundation, People First, and Innovation over Restraint, prioritizing developmental gains while mitigating risks like algorithmic bias and deepfakes. This “techno-legal” approach avoids a single restrictive AI law. 
• The White Paper 2026: The White Paper on AI Governance, released by the Office of the Principal Scientific Adviser on 24 January 2026, formalizes a shift from “command-and-control” to “Governance-by-Design.” This model mandates that ethical guardrails and legal compliance be embedded directly into the AI architecture at the code level. It adopts a risk-proportionate mechanism where high-impact systems in healthcare and law enforcement face mandatory human oversight to prevent autonomous systemic bias.