Voice AI moves beyond scripts as Indian firms tap multilingual bots

Planning a holiday, booking a train ticket, or checking a loan balance is no longer about pressing keypad numbers or waiting for a human agent. AI-powered voice assistants are now handling these conversations in real time, across multiple Indian languages, and with far greater accuracy.

Travel platform MakeMyTrip’s Myra is one example. A family can ask the bot in Hinglish to plan a Paris trip, and it will respond instantly with flights, hotels, and itineraries. “It’s a significant shift from the rigid, script-driven systems of the past,” Sanjay Mohan, group CTO at MakeMyTrip, told TOI's Shelley Singh.

Earlier bots relied on fixed natural language processing (NLP) scripts. If a customer said X, the bot replied with Y. “They were rigid, lifeless. Interrupt the bot, and it would just keep talking, oblivious to you,” said Ganesh Gopalan, cofounder of Gnani.ai. He recalled cases where a loan recovery bot kept asking for payments even when a customer said they were in a hospital.

The arrival of large language models (LLMs) and smaller domain-specific models (SLMs) has changed that. These engines can understand context, manage interruptions — called “barge-in” — and even switch between languages mid-sentence. Retailer Meesho, for example, now resolves over 60,000 daily customer queries through AI voice agents.

At the core of this shift are three technologies: automatic speech recognition (ASR) that converts speech to text, language models that generate context-aware responses, and text-to-speech (TTS) that delivers human-like replies. Companies are fine-tuning smaller open-source models for industry-specific data, reducing cost and latency.

Emotion is also programmable. A collections bot can sound firm, while a healthcare bot can offer empathy with lines such as: “Main aap ki pareshani samajh sakta hoon. Please tension mat lijiye.”

The biggest push is in vernacular adoption. India’s tier-2 and tier-3 towns, where English penetration is low, are now central to this growth. OpenAI’s GPT-5 has added support for 12 Indian languages, and rivals Anthropic and Perplexity are also investing in local capabilities.

Flipkart offers its app in 12 languages and uses voice AI for categories like beauty and customer service. “Users, especially in tier-3 markets, are shopping confidently because the app talks to them in their language,” said Sandhya Kapoor, SVP at Flipkart.

Microsoft Research India and ElevenLabs are also building tools for regional dialects, Hinglish, and slang. “The complexity is not just translation, but capturing dialects so AI responds naturally and contextually,” said Kalika Bali, principal researcher at Microsoft Research India.

For banks, auto firms, healthcare providers, and e-commerce platforms, these bots are now scheduling test drives, processing KYC, reminding EMI deadlines, and even giving medical advice — all in the customer’s preferred tongue.

The next stage of conversational AI in India may not emerge from urban metros but from users in Bihar, Bengal, or Punjab, speaking to a bot in their own dialect.

with TOI inputs