Gnani AI Launches Prisma v3.0 Speech Recognition for 10 Languages

29 Oct 2026, 5:50 pm · 2h ago · 1 min read · Analytics India MagazineNew

Bengaluru-based conversational AI startup Gnani AI has launched Prisma v3.0, an automatic speech recognition model supporting ten Indian languages alongside English. The model excels at transcribing code-mixed conversations, where speakers switch between Indian languages and English mid-sentence. On the Vistaar benchmark, Prisma v3.0 achieved a macro Word Error Rate of 16.93, outperforming alternatives from ElevenLabs, Microsoft Azure, and Deepgram. The platform offers four transcription modes—Verbatim, Transcribe, Code-mix, and Transliterate—and supports custom vocabulary prompting for up to 200 domain-specific terms, boosting terminology recall from 70.9% to 90.3% without requiring additional model fine-tuning.