Speechmatics Slashes Medical Transcription Errors by 40% with Swedish Model
Event summary
- Speechmatics launched a medical-grade Swedish speech-to-text model with a 3.91% keyword error rate (KWER), 40% lower than competitors.
- The model handles complex medical terminology, rapid multi-speaker dialogue, and diverse Nordic accents in noisy clinical environments.
- Speechmatics now supports dedicated medical models across Swedish, Finnish, Danish, and Norwegian.
- The model enables reliable automation in patient documentation, ambient scribes, and voice-driven workflows.
- Sully.ai, a partner, scaled from single-doctor clinics to enterprise customers with 500+ providers in under a year using Speechmatics' models.
The big picture
Speechmatics' new Swedish model addresses a critical gap in medical speech recognition, enabling more accurate and efficient documentation in Nordic healthcare. This launch comes as healthcare organizations increasingly adopt ambient documentation and autonomous AI agents, where transcription accuracy is essential. The model's success could set a new standard for medical speech recognition in the region, potentially influencing global standards.
What we're watching
- Adoption Pace
- How quickly Nordic healthcare providers will integrate the new Swedish model into their workflows.
- Competitive Response
- Whether competitors like OpenAI, AssemblyAI, Google, and Deepgram will respond with improved models.
- Automation Expansion
- The pace at which autonomous medical AI workflows will expand across the Nordics.
Related topics