MRI·CT Image Interpretation Recorded by Voice... Selvas AI Profitable for 4 Consecutive Years

Interview with Yoon Jae-sun, Head of Speech Recognition Business at Selvas AI
Development of 'Selvy MediVoice' Supporting Physician Interpretation Wording
Efforts to Enhance Accuracy and Security in Handling Health Information

"When a doctor interprets, a young junior doctor next to them has to type what is being said. There was a demand to introduce technology to assist with this typing task in hospitals, which led to the creation of this service."

Yoon Jae-seon, Head of Speech Recognition Business at Selvas AI, recently explained to reporters the reason behind developing their medical speech recognition service, 'Selvy MediVoice.' This service is the first in Korea to incorporate a speech recognition engine that supports recording CT, MRI, and other imaging interpretation results by voice instead of handwriting. Developed in collaboration with Severance Hospital over three years of joint research and development, this commercialized product is currently used at Samsung Seoul Hospital, CHA Hospital, Paik Hospital, and others.

Selvas AI is an AI company possessing human-computer interaction (HCI) technologies such as speech recognition, speech synthesis, and handwriting recognition. Last year, it achieved sales of 23.4 billion KRW and an operating profit of 1.3 billion KRW on a standalone basis, and it forecasts sales growth of over 25% this year. Yoon, the business head, is credited with leading the previously loss-making speech recognition division to four consecutive years of profitability.

Yoon Jae-sun, Head of Speech Recognition Business at Selvas AI, is being interviewed on the 13th at the Selvas AI conference room in Geumcheon-gu, Seoul. Photo by Jin-hyung Kang aymsdream@

In the medical field, speech recognition accuracy is crucial because it diagnoses patient conditions and records medical staff decisions. In medical settings, disease names or medical terms are often spoken in English, numbers in Korean, and formatting such as automatic paragraph breaks is also necessary. Yoon said, "Through engine upgrades, we improved accuracy from about 92% to 96%. Having stenographers within the hospital to refine data for training also greatly contributed to improving recognition rates."

Since patient information is involved, security is handled with great care. Selvas AI delivers cloud-based services to primary hospitals and either cloud-based or server-installed services to secondary and tertiary hospitals depending on demand. Yoon said, "For the cloud-based service, it operates within Amazon Private Zones. When information is sent to Amazon, it is encrypted, and the data is stored for about a month before being deleted."

Yoon Jae-sun, Head of Speech Recognition Business at Selvas AI, is being interviewed on the 13th at the Selvas AI conference room in Geumcheon-gu, Seoul. Photo by Kang Jin-hyung aymsdream@

Selvas AI is also conducting another proof of concept (PoC) project at Samsung Seoul Hospital to promote hospital smartification. Yoon said, "Regarding two-person face-to-face consultations, we developed a microphone that can separate speakers and a solution that converts this into meeting minutes." This service is expected to be used in various medical departments beyond radiology.

They plan to further improve hospital efficiency and reduce the time between diagnosis and treatment. Yoon said, "The module that acts as a gateway linking intensive care unit information such as oxygen saturation, blood pressure, and body temperature to the EMR is only available from our affiliate Mediana. We are also discussing a service that integrates speech recognition so nurses can register results in the EMR without typing directly."