As the artificial intelligence (AI) revolution sweeps the world, India is constrained by a familiar problem — that of language. The early adopters in India are tech-savvy, educated users comfortable in English. They type prompts into ChatGPT, Gemini, and Claude to generate answers, create presentations, write code and experiment with the latest AI tools. 

Beyond this visible layer of users lies a far bigger market — millions who want to interact via conversations, in Hindi, Tamil, Bengali, Marathi, Telugu, Kannada etc and dozens of dialects. For them, the future of AI may not arrive through text boxes and prompts, but through a simple question spoken into a phone. According to the Internet and Mobile Association of India (IAMAI) traditional English interfaces fail to reach the masses. Only around 10 per cent of users speak English. Voice allows access to over 90 per cent of the users who prefer regional Indic languages to search the web using their natural speech. This is extending to AI as well.

Ganesh Gopalan, co-founder and chief executive of voice AI company Gnani.ai says, “Text assumes literacy. Voice in a person’s own language removes both barriers — literacy and comfort with English.” Bengaluru-based Gnani.ai, part of the India AI mission, is developing sovereign voice AI models.

The next 300 million users

According to Navnit Nakra, partner and technology sector leader at PwC India, voice AI should be viewed not merely as a software category but as an inclusion platform. “The larger opportunity is voice as an access and workflow layer across banking, insurance, commerce, healthcare, education and citizen services,” says Nakra.

According to Nasscom, voice AI solutions could help bring the next 300 million Indians onto digital platforms, which is why, adds Nakra, “The voice AI opportunity should be viewed not only as a software market, but as an inclusion, productivity and services-transformation market.”

According to the Internet and Mobile Association of India (IAMAI), there are more than 548 million active rural internet users in India. Many access the internet primarily through smartphones. Yet navigating apps, filling forms or searching for information often remains cumbersome, particularly for first-generation AI users. Voice makes the access simple. Instead of typing a prompt  a user could simply ask in his language. Instead of navigating multiple menus, a patient could call and schedule an appointment in their local language. A farmer could check loan details or insurance status through a conversation rather than a screen full of text. Voice takes AI to the masses.

To sound human

The rise of voice AI is also creating a new generation of technology companies. For Davit Baghdasaryan, CEO and cofounder of Krisp, the journey began with a personal frustration. In 2017, while working at cloud communications company Twilio and travelling frequently between San Francisco and Armenia, he often found himself taking work calls from noisy cafés and public spaces.

“I wished there was a simple way to remove background noise and make conversations clearer,” he says. Together with co-founder Arto Minasyan, he built the first version of Krisp as an AI-powered noise-cancellation tool.

That product has evolved into a voice infrastructure company processing more than 80 billion minutes of voice conversations every month. Krisp’s technologies now include accent conversion, voice translation, conversational intelligence and infrastructure platforms that support AI agents. The company’s experience reflects a broader trend. As generative AI matures, the challenge is no longer producing answers. It is making conversations feel natural, or making machines sound human.

Most users assume voice AI is simply speech converted into text, processed by a language model and converted back into speech. The reality is considerably more complex. The hardest challenge is maintaining a real-time conversation. Unlike chat interfaces where users wait for responses, voice interactions require sub-second responsiveness. Delays longer than a second can make conversations feel awkward and robotic.

Baghdasaryan says, “Voice agent usage grew nine-fold in 2025. But most voice AI agents still fail in predictable ways once they leave demo environments.”

Background noise, interruptions, accents, poor network quality and overlapping conversations can reduce accuracy.

“Background voices push speech-to-text word error rate from 5 per cent to over 30 per cent, and voice activity detection misfires cause AI agents to ignore real interruptions or hallucinate them,” he adds.

Bengaluru-based Exotel, which processes more than 25 billion voice transactions annually, has developed systems that can distinguish between a caller who is thinking, someone who is speaking mid-sentence, and a network interruption. Its AI platform supports autonomous voice bots, real-time assistance for human agents and automated conversation analysis.

 Sachin Bhatia, cofounder and chief growth officer, Exotel says, “The hard problem in voice isn’t the model. It is a real-time conversation. We focus on solving this problem.” Most voice systems today follow a three-step process: Speech-to-text, language model processing and text-to-speech conversion.

Gnani.ai is building voice-to-voice architectures that remove text as an intermediary altogether. The goal is to reduce conversation latency below 500 milliseconds — fast enough for AI interactions to feel natural. Given that masses prefer voice interfaces, banks, insurers and lenders are using voice AI for customer onboarding, loan collections, payment reminders, verification processes and customer support.

 Gopalan of Gnani.ai says, “BFSI remains the strongest early adopter. Voice AI solves compliance and cost simultaneously, which is rare.” Gnani.ai’s customers include Axis Bank, Muthoot Finance, and Bank of Baroda.

 Healthcare is another fast-growing segment. Voice agents can schedule appointments, remind patients to take medication, conduct follow-ups and escalate concerns to healthcare professionals. In a country where doctor shortages remain acute, such systems could also help extend healthcare access at scale.

Commerce and logistics companies are also embracing the technology. Exotel’s client roster includes Uber, Swiggy, IKEA, Gojek and ICICI Bank. JSW MG Motor India uses Exotel’s solutions for automated feedback collection and customer engagement. According to the company, the initiative has improved customer satisfaction scores by 22 per cent. Ecommerce company Myntra uses voice AI to deliver personalised customer interactions, while Piramal Finance employs conversational AI across collection and customer engagement workflows.

PwC estimates that AI-enabled voice systems can reduce customer service costs by as much as 70  per cent, while improving first-call resolution rates and lowering escalation volumes. The attraction is obvious: Millions of repetitive customer interactions can be handled more efficiently and in multiple languages.

If India is the world’s toughest market for voice AI, it is because of its linguistic complexity.

People routinely switch between languages within a single sentence. Accents vary across regions.

Conversations often occur on low-quality phone connections amid background noise. Global speech models, many trained primarily on Western datasets, often struggle in such environments.

To address this, voice AI companies are building models specifically designed for local needs. Exotel supports more than 15 Indian languages. Gnani.ai claims its systems have been trained on millions of hours of real-world audio and now support over 40 languages. The focus is increasingly on code-switching—the uniquely Indian habit of mixing languages seamlessly within conversations.

That capability could prove crucial for bringing voice AI to rural India. “Loan reminders, balance queries and renewal confirmations in a user’s own dialect remove one of the biggest barriers to participating in the formal economy,” says Gopalan. As adoption grows, companies are experimenting with multiple monetisation models. These include subscriptions, enterprise licensing, call volumes, transaction volumes and AI usage. Newer models tie revenue directly to outcomes. Instead of charging for minutes, providers may charge based on successful collections, completed transactions or customer service resolutions.

Nakra says, “Voice platforms should not only answer queries. They should listen, reason and act. The real shift will come from monetising ‘voice automation’ to ‘voice intelligence’. Business will be tied to measurable outcomes like actions completed or issues resolved.”

Whose voice?

The rise of voice AI also introduces new risks, like voice cloning. Advances in synthetic speech mean AI can now recreate a person’s voice with startling accuracy. While technology can improve customer experiences and accessibility, it also raises concerns around fraud, impersonation and misinformation.

“If customers cannot trust whether a voice interaction is genuine, adoption will suffer,” says Nakra. Industry leaders argue that safeguards will be essential.

These include explicit user consent, synthetic voice disclosures, authentication systems, watermarking technologies, audit trails and human oversight for sensitive transactions.

Bias is another concern. Voice systems must work equally well across genders, age groups, accents and socio-economic backgrounds. Companies are investing in diverse datasets and continuous testing to ensure accuracy across different user populations. The challenge is ongoing rather than solved.

The global AI industry’s attention today remains focused on large language models and chat interfaces. Yet India’s experience may point to a different future. For much of the world, AI is still something people type into. For millions of Indians, it may become something they simply talk to. The implications extend far beyond technology. Voice could make banking more accessible, healthcare more reachable, education more personalised and government services easier to navigate. Though voice AI won’t solve every challenge. Literacy gaps, trust issues, authentication requirements and infrastructure could create hiccups in adoption.

But as multilingual models improve and conversational systems become more reliable, voice will emerge as one of the most powerful enablers of AI inclusion. The next chapter of synthetic intelligence may not be written. It may well be spoken.

Why India needs voice AI

Mass appeal: Around 65% of India’s internet users rely on voice search, generating an estimated 1.2 billion monthly queries.Without voice interface, internet access would have been limited. AI also won’t reach the masses without voice support

Digital inclusion: Voice search lowers barriers for users with limited literacy or typing skills

Multilingual inclusion: With 22 official languages and multiple dialects, voice AI can cut across linguistic barriers far more effectively than text

Rural commerce enabler: Farmers, shopkeepers, and small businesses in semi-urban and rural  areas can transact, search, and communicate using spoken commands

Greater access: Voice AI can increase access to government, healthcare, financial service 

Global competitiveness: Developing robust Indic voice AI models positions India as a leader in inclusive AI

Risks & challenges

Accuracy in dialects: Indian languages have complex variations; misrecognition can  exclude users

Privacy concerns:Voice data collection raises questions about consent and misuse 

Voice phishing: Voice cloning could amplify risks of impersonation, thefts 

Source: Industry, stateglobe.com