Granted Patent B2

US 12,733,872 · App. 18/522,320 · Granted Sep 15, 2026

Inventors: Seong Ji Kang (Seoul, KR); Hye Kang Roh (Seoul, KR); Joo Young Kim (Seoul, KR); Do Hyun Lee (Gyeonggi-do, KR); Hwa Young Jeong (Incheon, KR)

Assignee: WELT CORP., LTD

A61B5/4803G10L15/22G10L25/66

Abstract

Claims (18)

1 . A computer-implemented method of training at least two AI (artificial intelligence) engines for performing disease management using voice data, wherein the at least two AI engines comprise a common AI engine and an individual AI engine, the method comprising:

collecting user's voice data;

preprocessing the user's voice data;

extracting context data and out-of-context data from the voice data, wherein

the context data is characteristic data generated on a context including sentence completeness, word composition, or a vocabulary, and

the out-of-context data is characteristic data generated out of the context including a tone, a pitch, or a stuttering level;

extracting a plurality of first lower context data from the context data based on a first reference lower context data;

extracting a plurality of first lower out-of-context data from the out-of-context data based on a first reference lower out-of-context data;

training the common AI engine using the plurality of first lower context data and the plurality of first lower out-of-context data;

extracting a plurality of second lower context data from the context data based on a second reference lower context data;

extracting a plurality of second lower out-of-context data from the out-of-context data based on a second reference lower out-of-context data; and

training the individual AI engine using the plurality of second lower context data and the plurality of second lower out-of-context data,

wherein

the first reference lower context data is a data appears, with a predetermined ratio including percentage of eighty or more, in users having a specific disease,

the second reference lower context data is a data observed in a specific user or a specific user group having the specific disease, excluding the first reference lower context data,

the first reference lower out-of-context data is a data appears, with the predetermined ratio including percentage of eighty or more, in the users having the specific disease,

the second reference lower out-of-context data is a data observed in the specific user or the specific user group having the specific disease, excluding the first reference lower out-of-context data, and

the first reference lower context data and the first reference lower out-of-context data are adaptively changed according to an accumulation of the user's voice data.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (26)