Shunyalabs.ai has launched ZeroMed, a next-generation automatic speech recognition (ASR) system designed specifically for healthcare environments. The company calls it “a breakthrough domain-optimized automatic speech recognition system tailored to medical and clinical workflows,” delivering “best-in-class accuracy and ultra-low training overhead.”
ZeroMed reportedly achieves a word error rate (WER) of 11.1% and a character error rate (CER) of 5.1%, outperforming leading systems such as Whisper V3, ElevenLabs Scribe, Gemini 2.5 Flash, and AWS Transcribe. Its standout feature: it reaches “full convergence in just 3 days of training on 2 × A100 GPUs” which is a fraction of the time competitors require.
That speed could make it easier for health systems to keep the model updated with the latest medical terminology, procedures, and drug names, all areas where general-purpose speech models often fall behind.
Built for the Realities of Clinical Conversation
Medical speech recognition is notoriously challenging, filled with acronyms, rapid exchanges, and specialized language. According to Shunya Labs, ZeroMed tackles these issues through “domain-aware vocabulary and formatting,” “robust speaker diarization and context tracking,” and “accent-robust, low-domain bias.”
The model can distinguish between multiple speakers, such as a clinician, patient, or caregiver, even during overlapping dialogue. It supports ICD and LOINC codes, dosage normalization, and abbreviation expansion, helping reduce manual editing and transcription errors.
Equally important for clinical use, ZeroMed’s “real-time-first architecture” means it “delivers identical recognition quality in both real-time and offline modes,” enabling it to power live scribing, dictation, and telemedicine documentation without lag.
Privacy-First Deployment for Regulated Environments
While many AI transcription tools rely on the cloud, ZeroMed was built for privacy-sensitive healthcare systems. Shunya Labs emphasizes that the model “can run on-premises on CPU-only servers (no cloud dependency), providing full data control and compliance with healthcare privacy standards (HIPAA, GDPR, etc.).”
That on-premises option gives hospitals and enterprises full control over patient data while maintaining performance parity with GPU and cloud deployments, a rare combination in medical AI tools.
Industry-Leading Performance
Benchmark tests released by Shunya Labs show ZeroMed leading across accuracy, efficiency, and deployment flexibility, outperforming other medical ASR systems that require weeks of training and GPU-only setups.
“At Shunya Labs, we believe medical transcription must be not just fast, but flawlessly accurate — every dosage, diagnosis, and timestamp matters. ZeroMed embodies that vision. We’ve reduced the cost and time to train, making high-fidelity ASR accessible to more healthcare systems,” said Ritu Mehrotra, CEO and Founder of Shunyalabs.ai.
CTO Sourav Banerjee added, “Our goal with ZeroMed wasn’t incremental improvement — it was to redefine medical speech recognition: fewer corrections, lower latency, and complete data privacy.”
Availability and What’s Next
ZeroMed is now available for preview and pilot evaluation by healthcare and healthtech organizations. Shunya Labs is onboarding early partners for integration and feedback, with on-prem CPU-only options available for strict compliance environments.
Currently available in English, ZeroMed will soon expand to Indian and other international languages, reflecting the company’s global ambitions. Interested organizations can request a demo or pilot access at www.shunyalabs.ai/ZeroMed.
As healthcare systems search for ways to reduce administrative burden and clinician burnout, tools like ZeroMed could signal a shift toward AI-driven documentation that’s both accurate and privacy-compliant. With its ability to train fast, run locally, and adapt to evolving vocabularies, Shunya Labs’ approach stands out in an increasingly competitive field of speech AI.
