Enterprise voice AI startup Gnani.ai has released Prisma v2.5, a speech-to-text model tailored to India's complex linguistic environment with support for 12 languages and dialects. The model aims to improve transcription accuracy and responsiveness for critical industry uses while promoting sovereign AI capabilities.
- Prisma v2.5 supports 12 Indian languages and dialect variations.
- Model improves accuracy in noisy, multilingual, real-world voice environments.
- Hosted locally in India for faster, low-latency voice applications.
What happened
Gnani.ai, an Indian enterprise voice AI startup, has introduced Prisma v2.5, a new speech-to-text model designed specifically for India's diverse linguistic landscape. The model supports 12 languages and incorporates natural code-switching, ambient noise, and dialect variations into its training, drawn from 14 million hours of proprietary Indic speech data.
Prisma v2.5 is accessible via APIs and optimized for use cases in industries such as banking, insurance, and healthcare where transcription accuracy is critical. The model has demonstrated better error rates and lower latency than global and domestic competitors by leveraging local data centers, enabling improved real-time performance in telephony and voice applications.
Why it matters
India’s voice-first market presents unique challenges due to multilingualism, accent diversity, and network constraints that traditional speech recognition models—often built for ideal studio conditions—fail to address effectively. Prisma v2.5’s tailored approach fills this gap by training on real-world Indian voice data, making it highly relevant for local enterprises handling critical voice data.
The launch contributes to India’s growing sovereign AI ecosystem, aiming to reduce dependence on global AI providers. Gnani.ai’s focus on localization and government backing under the IndiaAI Mission positions it strongly against international players entering the Indian market, supporting national strategic interests in AI technology sovereignty.
What to watch next
Gnani.ai plans to expand Prisma’s footprint beyond India, targeting markets like Japan, the Philippines, and the Middle East with similarly localized voice models. The company is investing its recent $10 million Series B funding primarily in research, talent acquisition, and entering new verticals and international markets.
Competition in India’s sovereign voice AI space is intensifying with startups such as Sarvam AI and BharatGen also developing dedicated models. Observers should watch how these homegrown solutions compete against established global providers in accuracy, performance, and enterprise adoption, especially amidst evolving geopolitical restrictions on advanced AI model exports.