Gnani AI has released Prisma v3.0, a speech-to-text model built for the unique complexities of Indian telephony. The update aims to improve transcription accuracy in code-mixed regional languages, helping enterprises in the banking and service sectors better manage high volumes of customer voice data.
Gnani AI has launched Prisma v3.0, a specialized speech-to-text model designed to handle the linguistic nuances of the Indian market. The model is specifically engineered to address the challenges of Indian telephony, where speakers often switch between regional languages and English in a single conversation. This phenomenon, known as code-mixing, has historically been difficult for global AI models to transcribe with high accuracy.
Targeting Indian Dialects and Telephony
Unlike general-purpose models that are trained on broad, international datasets, Prisma v3.0 was trained on localized telephone audio. This focus helps the system filter out background noise and recognize regional dialects, which are common in Indian call center environments. According to the company, the model uses a specialized scoring protocol to achieve a macro word error rate of 16.9 percent across nine Indian languages. This is an area where international competitors often struggle, as their models may not be optimized for the specific cadence or vocabulary of Indian regional speech.
Customization for Enterprise Needs
Enterprise clients, particularly in the banking, insurance, and customer support sectors, often require different types of data output from the same voice interaction. Prisma v3.0 introduces four distinct processing modes to address these needs. The Verbatim mode captures filler words for compliance and legal audits, where every detail is required. The Transcribe mode generates clean, normalized text suitable for database logging. The Code-mix mode preserves the original language script, while the Transliterate mode converts regional speech into English characters. This modularity allows companies to integrate the technology into different business functions without needing separate models for each task.
Industry Impact and Market Context
Gnani AI currently processes approximately 50 million voice interactions daily, and this update represents a push to solidify its footprint in the enterprise service sector. For Indian businesses, the ability to accurately transcribe calls is essential for automated customer support, regulatory reporting, and internal quality assurance. By allowing the input of up to 200 custom keywords per request, the platform also helps businesses ensure that company-specific terminology and product names are captured correctly.
While the technology offers clear operational benefits, the key monitorable for the industry will be the adoption speed among large-scale enterprise clients who currently rely on legacy systems or established global AI providers. Additionally, as automated processing of voice data grows, companies will continue to face strict data privacy requirements under Indian regulations. Investors in the broader Indian enterprise tech and software space may watch how such localized AI innovations influence the competitive landscape against larger global tech giants that are also competing for the Indian enterprise market.