Gnani AI Launches Prisma v3.0 for Mixed-Language Speech

Bengaluru-based Gnani AI has unveiled *Prisma v3.0*, a speech recognition model built to handle the messy reality of Indian conversations, where speakers frequently switch between English and Indian languages mid-sentence. The model supports ten languages, including Hindi, Bengali, Marathi, Gujarati, Punjabi, Kannada, Telugu, Tamil and Malayalam alongside English.

On the Vistaar benchmark, Prisma v3.0 posted a macro Word Error Rate of 16.93, outperforming competitors such as ElevenLabs Scribe v2, Microsoft Azure Speech-to-Text and Deepgram Nova 3. The company said it developed the model because, in its words, *"a correct transcript still wasn't a usable one"*, pointing to persistent issues with spoken numbers, brand names and short responses in enterprise voice systems.

The release introduces four transcription modes, Verbatim, Transcribe, Code-mix and Transliterate, giving developers flexibility in how mixed-language speech is rendered. Gnani also highlighted improvements in custom vocabulary recognition, language identification accuracy and reduced hallucinated text from noise-only audio, positioning Prisma v3.0 as a more reliable option for multilingual voice applications across telephony and VoIP systems.