Deepgram has introduced Nova-3 Medical, an AI-driven speech-to-text model designed to enhance medical transcription accuracy. As healthcare relies more on digital records and telemedicine, traditional transcription tools often struggle with complex medical terminology, leading to errors.
Nova-3 Medical addresses this challenge by using advanced machine learning and specialized medical vocabulary training to accurately transcribe medical terms and acronyms, even in difficult audio conditions. This ensures seamless integration with clinical workflows and electronic health records (EHRs), improving efficiency and patient care.
Deepgram’s CEO, Scott Stephenson, emphasized that Nova-3 Medical represents a significant advancement in AI-powered clinical documentation. The model allows developers to customize it for various medical specialties with up to 100 key medical terms, making it adaptable for different healthcare settings. Its enterprise-grade security features, including on-premises and Virtual Private Cloud (VPC) options, ensure compliance with UK data protection regulations and HIPAA standards.
Benchmarking results highlight Nova-3 Medical’s superior performance. It achieves a median Word Error Rate (WER) of 3.45%, reducing transcription errors by 63.6% compared to competitors. Additionally, its Keyword Error Rate (KER) of 6.79% ensures critical medical terms are accurately transcribed, minimizing miscommunication and improving patient safety. The model also excels in speed, transcribing speech up to 40 times faster than alternative solutions, making it ideal for real-time applications like telemedicine.
Nova-3 Medical is also cost-effective, with pricing starting at $0.0077 per minute of streaming audio—more than twice as affordable as leading cloud providers. This affordability allows healthcare tech companies to invest more in innovation. Deepgram’s latest AI model is set to revolutionize medical transcription, empowering developers to create more accurate, efficient, and cost-effective healthcare solutions.
