This speech recognition model specialized in medical vocabulary greatly reduces the error rate on the names of drugs, diagnoses and clinical procedures.
Deepgram has launched Nova-3 Medical, a variant of its Nova-3 model refined on clinical audio corpora to better recognize medical terminology — prescriptions, diagnoses, acts, anatomical terms. The model achieves a median error rate of 3.44%, presented as an improvement of more than 60% over the top competitor, with an error rate on key terms reduced by more than 40%.
Designed for clinical documentation, telemedicine and integration with electronic patient records, the tool also allows personalization by adding terms specific to each institution, with speaker identification and automatic formatting, in an architecture that complies with the confidentiality requirements of the healthcare sector.
Read the full article on the original source
Go to the source →