AI speech model cuts healthcare transcription errors

Deepgram has unveiled Nova-3 Medical, an AI speech-to-text (STT) model tailored for transcription in the demanding environment of healthcare.

Designed to integrate seamlessly with existing clinical workflows, Nova-3 Medical aims to address the growing need for accurate and efficient transcription in the UK’s public NHS and private healthcare landscape.

As electronic health records (EHRs), telemedicine, and digital health platforms become increasingly prevalent, the demand for reliable AI-powered transcription has never been higher. However, traditional speech-to-text models often struggle with the complex and specialised vocabulary used in clinical settings, leading to errors and “hallucinations” that can compromise patient care.

Deepgram’s Nova-3 Medical is engineered to overcome these challenges. The model leverages advanced machine learning and specialised medical vocabulary training to accurately capture medical terms, acronyms, and clinical jargon—even in challenging audio conditions. This is particularly crucial in environments where healthcare professionals may move away from recording devices.

“Nova‑3 Medical represents a significant leap forward in our commitment to transforming clinical documentation through AI,” said Scott Stephenson, CEO of Deepgram. “By addressing the nuances of clinical language and offering unprecedented customisation, we are empowering developers to build products that improve patient care and operational efficiency.”

One of the key features of the model is its ability to deliver structured transcriptions that integrate seamlessly with clinical workflows and EHR systems, ensuring vital patient data is accurately organised and readily accessible. The model also offers flexible, self-service customisation, including Keyterm Prompting for up to 100 key terms, allowing developers to tailor the solution to the unique needs of various medical specialties.

Versatile deployment options – including on-premises and Virtual Private Cloud (VPC) configurations – ensure enterprise-grade security and HIPAA compliance, which is crucial for meeting UK data protection regulations.

“Speech-to-text for enterprise use cases is not trivial, and there is a fundamental difference between voice AI platforms designed for enterprise use cases vs entertainment use cases,” said Kevin Fredrick, Managing Partner at OneReach.ai. “Deepgram’s Nova-3 model and Nova-3-Medical model, are leading voice AI offerings, including TTS, in terms of the accuracy, latency, efficiency, and scalability required for enterprise use cases.”

Benchmarking Nova-3 Medical: Accuracy, speed, and efficiency

Deepgram has conducted benchmarking to demonstrate the performance of Nova-3 Medical. The model claims to deliver industry-leading transcription accuracy, optimising both overall word recognition and critical medical term accuracy.

• Word Error Rate (WER): With a median WER of 3.45%, Nova-3 Medical outperforms competitors, achieving a 63.6% reduction in errors compared to the next best competitor. This enhanced precision minimises manual corrections and streamlines workflows.
• Keyword Error Rate (KER): Crucially, Nova-3 Medical achieves a KER of 6.79%, marking a 40.35% reduction in errors compared to the next best competitor. This ensures that critical medical terms – such as drug names and conditions – are accurately transcribed, reducing the risk of miscommunication and patient safety issues.

In addition to accuracy, Nova-3 Medical excels in real-time applications. The model transcribes speech 5-40x faster than many alternative speech recognition vendors, making it ideal for telemedicine and digital health platforms. Its scalable architecture ensures high performance even as transcription volumes increase.

Furthermore, Nova-3 Medical is designed to be cost-effective. Starting at $0.0077 per minute of streaming audio – which Deepgram claims is more than twice as affordable as leading cloud providers – it allows healthcare tech companies to reinvest in innovation and accelerate product development.

Deepgram’s Nova-3 Medical aims to empower developers to build transformative medical transcription applications, driving exceptional outcomes across healthcare.

(Photo by Alexander Sinn)

See also: Autoscience Carl: The first AI scientist writing peer-reviewed papers

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with other leading events including Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.