Deepgram has added a speech‑to‑text (STT) artificial intelligence (AI) model specifically designed for the health care sector that needs to find ways to reduce costs and increase efficiency.
The Nova‑3 Medical model is based on a latent space architecture that compresses representation of data points in a way that preserves only the essential features needed to inform the input data’s underlying structure. The approach makes it possible to encode complex speech patterns into a highly efficient representation in real time while providing higher levels of accuracy.
Rather than simply providing access to an STT model, however, Deepgram is going a step further by training models that are optimized for use cases that are unique to a vertical industry sector, says Deepgram CEO Scott Stephenson.
For example, rather than requiring someone to manually take notes during a patient examination, the Nova-3 model will capture voice interactions, including drug names, conditions and procedures, to automatically create a transcription, he adds.
With a median Word Error Rate (WER) of 3.45%, Nova-3 Medical can reduce errors by as much as 63.6% compared to rival platforms, the company claims.
Additionally, Deepgram claims Nova‑3 Medical transcribes speech 5 to 40 times faster than most other platforms, which is critical for telemedicine services. That capability provides the health care sector with an AI model that understands clinical terminology versus requiring the healthcare organization to train and custom a general-purpose AI model such as ChatGPT, notes Stephenson. Health care organizations can, however, still customize the outputs provided using a set of prompts based on 100 terms that Deepgram provides.
Transcriptions can also be more easily integrated into workflows based on electronic health care records (EHR) to eliminate errors that might occur when data is entered manually, says Stephenson.
Finally, Deepgram provides health care organizations with the ability to deploy these AI models in an on-premises or cloud computing environment as needed, to comply with various regulations, he notes.
Pricing, meanwhile, starts at $0.0077 per minute of streaming audio as part of an effort to make Nova-3 Medical broadly accessible to as many health care providers as possible, Stephenson says.
Once considered laggards when it comes to adoption of emerging technologies, the health care industry is emerging as an early AI adopter largely out of necessity. Most health care providers operate on thin margins and many of them can’t find, much less, afford to dedicate resources to data entry. “Health care went from being conservative to very bullish,” says Stephenson.
It’s not clear yet what impact adoption of AI might have on the quality of patient care, but in theory at least every minute a health care professional isn’t entering data is clearly a potential opportunity to improve the care provided by medical professionals that are generally being overwhelmed by data.
The challenge, of course, then becomes not just finding a place to store all that data but also analyzing it in a way that ultimately results in better outcomes, not just for the patient but also everyone involved in caring for them.
