Oriserve launches Whisper – Hindi to English Apex, to power next gen AI for India’s multi-lingual business needs

Whisper is a state-of-the-art automatic speech recognition (ASR) model, trained on a large and diverse data set of multilingual and multi task audio data

Oriserve, a leading generative AI platform transforming enterprise communication, announced the Open-Source launch of Whisper – Hindi2Hinglish-Apex – its most advanced Speech-to-text (STT) model for conversational Hindi, Hinglish, and Indian accented English. This model, available on the Hugging face platform, marks a giant leap in Oriserve’s mission to make AI accessible, inclusive, and truly represent the diverse linguistic needs of the Indian market.

Bridging the English only AI disconnect: Curating AI for India

India’s linguistic diversity has always been a cultural strength and a technological challenge. The dominance of English-based AI models has excluded millions of regional users from fully engaging with digital platforms or truly leveraging the real power of AI innovations. And although Speech-to-text (STT) technology has rapidly evolved in the last decade, most large-scale STT models – such as Google speech, Amazon transcribe, or OpenAI’s Whisper, continue to struggle to handle India’s linguistic diversity and accent variations. In India, where languages and dialects change every few hundred kilometres, curating a speech-to-text for India poses a critical challenge.

Whisper is a state-of-the-art automatic speech recognition (ASR) model, trained on a large and diverse data set of multilingual and multi task audio data. However, it operates best only on clean and standardised dataset. When met with telephonic audio with background noise or audio with strong regional accent - which is the case in majority of Indian call scenarios, Whisper’s existing infrastructure is known to underperform. And this is where Oriserve steps in!

Building on Whisper’s robust foundation, Oriserve introduced the Whisper-Hindi2Hinglish-Apex - a fine-tuned model that is optimised for recognising Hindi, Hinglish, and Indian accented English, specially addressing challenges in regional accent recognition and real-life call quality conditions. Trained using over 1000 plus hours of recorded audio data, Whisper-Hindi2Hinglish-Apex is a notch higher, combining accuracy, performance, and contextual understanding of real-world Indian use cases.

Speaking about this innovation, Anurag Jain, Co-Founder, Oriserve, said, “Whisper-Hindi2Hinglish-Apex is a small step in making AI truly inclusive for India’s diverse linguistic ecosystem. And by open-sourcing this model, we aim to empower the wider community to innovate, localise, and redefine how Voice AI understands India. For small businesses, especially in tier 2 and tier 3 towns, it means faster digital adoption and better customer engagement. And for larger businesses, it is an opportunity to unlock the vast potential of India’s thriving regional markets by enhancing regional customer engagement and delight!”  

Adding to this, Maaz Ansari, Co-Founder, Oriserve, said, “This launch is as much about creating an inclusive AI speech model as it is about democratising advanced AI Voice technology. Whether it is a start up in a tier 3 town or a large multi-national expanding its footprints into regional India, we want every business to be able to speak the language of its customers! And with this goal, we aim to continue innovating and creating meaningful AI Voice tools across Indian languages, in the near future.”

The Whiper-Hindi2Hinglish-Apex by Oriserve, comprises of over 800 million parameters, balancing compact design and cutting-edge tech, and boasts of:

-          8x faster interface than larger models while maintaining equivalent accuracy

-          42% higher average performance over Whisper’s pre-trained baseline

-          Enhanced robustness in accented, hybrid and noisy conversational audio

For Businesses, this means the ability to deliver voice-led experiences in the customer’s preferred language, enhancing trust, accessibility, and overall engagement. Enterprises across BFSI, Telecom, Healthcare, Education and private services, can now have access to sharper analytics, faster automation, and smarter conversational AI, all while reducing dependency on foreign, expensive, cloud only models. For citizens and small enterprises across tier 2 and tier 3 towns, this represents a step towards digital inclusion, enabling them to interact with technology in a confident manner.  

The Road Ahead: Building a Multilingual AI for a Billion Voices

Whisper-Hindi2Hinglish-Apex is the third in Oriserves Open-Source series and the company aims to expand this model to include several other regional languages, including Marathi, Gujarati, Tamil, Telugu, Kannada, Malayalam, Bengali, and Punjabi, paving the way for a unified, multilingual AI ecosystem.