Introducing Nova-3 Medical Multilingual: Medical Speech Recognition in 10 Languages

Deepgram has expanded its Nova-3 Medical speech recognition platform with a multilingual model supporting Dutch, English, French, German, Hindi, Italian, Japanese, Portuguese, Russian, and Spanish, including batch and real-time transcription and automatic switching between supported languages. Designed for medical vocabulary rather than general multilingual speech, the model showed lower average word error rates than Deepgram’s general multilingual model, declining from 6.62% to 4.96% in batch transcription and from 8.32% to 5.86% in streaming tests across more than 115 hours of curated medical audio. It also reduced errors involving clinically important entities such as doses, conditions, procedures, and drugs, with batch entity error rates falling by 37% and streaming rates by 40%. Deepgram reports that the model achieved the lowest average batch word error rate among systems tested across all 10 languages and supported applications including ambient clinical documentation, voice agents, and clinician dictation. The company also upgraded its English-only Nova-3 Medical model, reporting improved transcription and entity accuracy, while directing English-only users to the English setting and multilingual or code-switching use cases to the new multi-language setting.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.