.webp)
.webp)
ConvoZen has launched an end-to-end conversational AI platform and two speech models designed for India's multilingual, voice-led customer interactions.
It announced the products at its ConvoZen Conversational AI Summit in Bengaluru, where executives from technology firms, financial services groups and consumer brands discussed how organisations are deploying AI in customer operations.
The releases centre on a single platform that combines customer-facing conversational agents with tools for human agents and supervisors. ConvoZen also introduced two in-house speech models: Akshara for speech-to-text and Ragini for text-to-speech. It describes both as "indigenous frontier" models trained for Indian languages and real-world telephony audio.
ConvoZen emerged from NoBroker, the property services business, and positions its platform as an integrated alternative to point conversational AI tools. Many enterprises use separate products for chatbots, voice bots, agent assist and analytics; ConvoZen aims to provide a single stack that covers the full lifecycle of customer interactions.
Four agent layers
The platform is organised into four layers. Conversational AI Agents handle interactions across voice, chat and WhatsApp in multiple languages. Copilot AI Agents provide real-time assistance for customer service staff. Supervisor AI Agents monitor conversations for quality and compliance. Customer 360 AI Agents consolidate customer intelligence into a unified view.
ConvoZen says it serves more than 50 enterprise customers, including TATA AIG, HDFC Life, CARS24, Spinny, Pilgrim, The Souled Store, Maruti Suzuki, Kotak AMC and Jana Bank.
The company also cited the influence of Nandan Nilekani on its product direction, saying its roadmap has benefited from his guidance on building scalable digital infrastructure for India.
Telephony speech
Akshara is ConvoZen's speech-to-text model, trained on anonymised real-world business-to-consumer telephonic conversations. It is built for 8 kHz call recordings and short, variable consumer speech, and targets multi-dialect and code-switched conversations, as well as business terminology and contextual phrasing.
ConvoZen also made comparative performance claims against other automatic speech recognition systems. It says Akshara achieved a 32% reduction in errors compared with the next-best model it tested, and more than 55% fewer errors than leading global alternatives on public benchmarks including IndicVoices. It also says Akshara outperformed Indian and global models across nine regional languages on public benchmarks such as Indic Voices and Swara.
Alongside Akshara, ConvoZen introduced the Indic Conversational AI Voice Benchmark, which it described as an open benchmark focused on authentic business-to-consumer telephonic interactions. The company says Akshara leads performance on that benchmark by a significant margin.
Text to speech
Ragini is the firm's text-to-speech model, trained on proprietary conversational data. It is designed for multilingual output and code-switching, and aims to improve pronunciation for Indian names, localities and commonly used terms, including UPI and Aadhaar.
In blind comparative tests across six languages, ConvoZen says Ragini outperformed Indian and global providers on naturalness and pronunciation accuracy. Ragini launches with multiple voices for deployments in banking and insurance, automotive, healthcare, direct-to-consumer retail and edtech.
Demand for speech systems tuned to India's languages and call-centre audio has increased as companies automate customer contact. Many widely used models are trained on studio-quality speech or datasets that do not reflect noisy telephone conditions and code-switching between English and local languages. ConvoZen is positioning Akshara and Ragini as alternatives for customer service operations that need higher accuracy on live calls.
Akhil Gupta, Founder, ConvoZen.AI & NoBroker.com, said, "We believe the future of customer operations lies in the coexistence of human and AI agents. Our unified platform brings together Conversational AI Agents across voice, chat, and WhatsApp with memory. India is a voice-first nation. Yet enterprise-grade voice understanding models trained on real Indian conversational data have been scarce.
"With Akshara and Ragini, we are introducing indigenous frontier speech models built specifically for India's multilingual, multi-dialect ecosystem. Our vision is to make AI truly conversational for Bharat, not just automated, but culturally and contextually aware."
Customer outcomes
ConvoZen cited several deployments. It says beauty and personal care brand Pilgrim has 65% of conversations handled by its Conversational AI Agents, alongside a 40% reduction in dependence on human agents and a customer satisfaction score of 4.25.
CARS24, the used-car marketplace, is using Supervisor Agents for quality audits. ConvoZen says this has fully automated quality audits and improved compliance processes.
Jana Bank has deployed conversational voice agents in sales workflows. ConvoZen reports overall sales growth of around 7% from the deployment, with improved conversion outcomes at scale.
NoBroker has used Supervisor AI Agents for automated call auditing and call-level diagnostic insights. ConvoZen reports up to a 25% improvement in call quality for NoBroker Interiors and an 8% uplift in site visits for NoBroker Builder, and says voice AI agents in the Builder business delivered around 5% incremental growth.
Summit speakers
The Bengaluru summit included an opening keynote by Sandeep Alur, CEO of Microsoft India. The agenda also featured a fireside chat with Pramod Varma, co-founder of EkStep, and panel sessions with speakers from Maruti Suzuki India, HDFC Life, Tata Tele, Meta, Niyo, Elevation Capital, SpeakX, Jana Bank, Pilgrim and The Souled Store. Ankit Bose, head of AI at NASSCOM, delivered the closing keynote.
ConvoZen says it will continue building telephony-optimised speech and a unified stack for customer operations as more enterprises adopt AI across service, sales and compliance workflows.