Shunya Labs Launches Pingala V1: Top-Ranked CPU-Based ASR Model Setting Performance Records

Precision-first speech AI for healthcare, defense and enterprise now available to companies worldwide

MENLO PARK, Calif., July 24, 2025 /PRNewswire/ --  Shunya Labs, the AI infrastructure company pioneering real-time, privacy-first systems for voice, language and reasoning, today announced the release of Pingala V1, its flagship automated speech recognition (ASR) model. The model sets new benchmarks in speech accuracy, multilingual support and speed, even though it was built on a $4 million budget.

Pingala V1 was originally developed to enhance mental health transcription, where phrases like "quetiapine" and "cutiepie" can lead to dangerous confusion. Now, it works across languages and accents in real time, linking speech to context, such as symptoms to diagnoses or customer complaints to service issues. It also remembers key details from earlier conversations. With under 250-millisecond latency, it's fast enough for ambulances, remote clinics, customer support centers and field operations.

"While trillion-dollar AI giants still struggle to distinguish between similar-sounding words, our team has built the world's most accurate speech recognition system," said Ritu Mehrotra, founder and CEO of Shunya Labs. "We didn't build it to chase leaderboards. We built it because patients were dying while waiting for doctors' notes to be transcribed accurately. Now, the model is ready to go live in hospitals, defense labs, call centers, and anywhere else where people can't afford to lose time."

Built for mission-critical sectors like healthcare, defense and enterprise, Pingala V1 delivers top-tier accuracy, flexible deployment and real-time multilingual support. Shunya Labs reports that the model achieved an industry-best word error rate of 2.94% across eight comprehensive benchmarks. This represents over 50% reduction of word error rate over the closest competitor. Unlike most other models, Pingala V1 does not need expensive GPUs or cloud infrastructure. It runs on regular CPUs, making it easier to deploy securely and affordably in sensitive environments.

Key Features of Pingala V1:

• Record accuracy: Lowest word error rates on major speech benchmarks.
• 200+ languages: Real-time multilingual translation and transcription, including within the same language and dialect switching.
• No special hardware: Runs on standard CPUs, no need for GPUs or cloud.
• Built for critical use: Optimized for healthcare, call centers and enterprise.
• Easy to deploy: Available via API, Docker, or edge-ready formats; SOC 2 and HIPAA compliant out of the box.

At a time when AI regulation, clinician burnout and infrastructure readiness are national priorities, Shunya Labs represents a new class of domain-specialized, precision-grade AI. Amid growing demand for private, explainable and low-latency AI, Pingala V1 delivers what Big Tech's black-box models can't.

About Shunya Labs

Shunya Labs builds real-time, privacy-first AI infrastructure optimized for voice, language and reasoning. Rooted in healthcare but designed for global deployment, Shunya's technology is redefining how machines understand human nuance, with products that are fast, frugal and fundamentally verifiable. Learn more at www.shunyalabs.ai.

Media Contact

Interdependence Public Relations
9193493786
[email protected] 

SOURCE Shunya Labs