Gnani.ai Releases Vachana STT Trained on 1M Hours of Voice Data
As voice becomes a primary interface for customer engagement, automation, and analytics in India, Bengaluru-based Gnani.ai has launched Vachana STT, a foundational Indic speech-to-text model trained on more than 1 million hours of real-world voice data. Released under the IndiaAI Mission, Vachana STT is positioned as core infrastructure rather than an application-layer add-on, aimed squarely at production-grade deployments across sectors.
A Foundation for VoiceOS
Vachana STT is the first public release in Gnani.ai’s forthcoming VoiceOS, a unified voice intelligence stack that brings together speech recognition, synthesis, understanding, and orchestration. Instead of stitching together multiple third-party services, the company says VoiceOS is built from first principles to function as a sovereign, end-to-end platform for enterprises operating at scale.
Built for India’s Real-World Audio
India’s speech landscape is notoriously complex: a multitude of languages and dialects, significant intra-regional accent shifts, variable audio quality across call centers and digital channels, and conversations that are rarely clean or scripted. Vachana STT targets this reality with proprietary multilingual datasets spanning 1,056 domains, designed to perform across noisy, omnichannel environments—without requiring additional fine-tuning.
According to Gnani.ai, this breadth of training helps maintain consistent accuracy where it matters most: automation workflows, compliance monitoring, agent assistance, and customer experience. The company frames the model as a critical layer of a broader infrastructure vision rather than a single-purpose API.
Accuracy and Language Coverage
Across evaluations on public datasets and live omnichannel audio, Gnani.ai reports:
- 30–40% lower word error rates (WER) on low-resource Indian languages
- 10–20% lower WER on the top eight Indian languages compared with leading providers
Languages covered include Hindi, Bengali, Gujarati, Marathi, Punjabi, Tamil, Telugu, Kannada, Malayalam, Odia, Assamese, and others. Detailed benchmarking and comparative reports are available to enterprises on request—a nod to buyers who increasingly expect transparent, domain-specific evaluation.
Latency, Scale, and Deployment
Vachana STT is already in production across BFSI, telecom, customer support, and large-scale voice automation systems, processing roughly 10 million calls per day. The platform supports both real-time and batch transcription via enterprise-grade APIs.
Latency remains crucial for use cases like live agent assist and compliance triggers. Gnani.ai reports p95 latency near 200 milliseconds under high concurrency—performance that opens doors to instant guidance, real-time analytics, and faster customer interactions. The model is also optimized for compressed audio from 8 kbps to 64 kbps, making it suitable for telephony as well as digital channels where network quality can vary.
Under the IndiaAI Mission
The release follows Gnani.ai’s selection under the IndiaAI Mission, a government initiative identifying a limited set of startups to build sovereign foundational AI models from India. The emphasis is on core AI infrastructure—particularly where datasets, language dynamics, and deployment conditions are uniquely Indian—rather than application-layer experiments alone.
“Speech recognition in India is not a localisation problem. It is a foundational systems problem,” said Ganesh Gopalan, co-founder and CEO of Gnani.ai. “Vachana STT is built as core infrastructure, trained on how India actually speaks, and designed to operate across channels, not just telephony. Being selected under the IndiaAI Mission reinforces our belief that foundational AI models must be built from India, with production reality at the centre.”
From Selective Localisation to Foundational Coverage
The focus on low-resource languages signals a shift away from narrow localisation toward broader foundational coverage. For enterprises, that means fewer edge cases and better resilience when real-world conditions deviate from the lab—especially in contact centers, field operations, and multilingual service lines where speech accuracy directly dictates outcomes.
Availability and Early Access
Vachana STT is available now via API for enterprise customers. Early adopters receive 100,000 free minutes of usage. Enterprises can request benchmarking data, technical evaluations, and API access directly from Gnani.ai.
As India accelerates toward voice-first experiences, Vachana STT—and the broader VoiceOS vision—aim to establish a durable layer of speech infrastructure built around India’s linguistic and operational realities. If the reported gains in accuracy and latency hold across production environments, this could be a pivotal step toward standardizing voice AI for the subcontinent’s most demanding use cases.
