[Life Chapter 3 Companies] 'Baikal AI' Diagnosing Dementia by Analyzing 'Speech Sounds'

CEO Yoon Ki-hyun of Baikal AI, a voice recognition AI company. Photo by Younghan Heo younghan@

[Asia Economy Reporter Eunju Lee] “An individual's speech contains important information that can identify risk factors for mental health issues such as dementia. Baikal AI has developed technology that analyzes the unique human speech patterns, which are difficult for AI to detect, to diagnose cognitive impairments and more.”

On the 17th, Asia Economy met with Yoon Ki-hyun, CEO of Baikal AI (56), at their office located in Jongno-gu, Seoul. Baikal AI is an AI technology-based startup founded in 2019, focusing on providing services that analyze Korean speech to offer information about individuals' mental health conditions such as dementia and depression.

CEO Yoon, who founded the company, has a 30-year career as a software developer. In 2018, he developed Hana Bank’s AI chatbot (Hi Chatbot) at the web AI company MindsLab. The chatbot he developed was the first in Korea to enable financial transactions. When users input speech, the chatbot recognized it and guided them through financial transactions. While he took considerable pride in this achievement, it also became a moment where he confronted the limitations of AI technology.

The chatbot had limitations in accurately recognizing and analyzing spoken language. He said, “The chatbot, which usually understood my voice well, could not recognize my voice after I had been drinking (when my pronunciation changed). I realized that technology to grasp the subtle changes in people’s speech, which inevitably vary due to many factors, was progressing slowly.” He then came to believe that there are many services that could be provided through technology that analyzes even the subtle changes in speech. For AI technology to evolve in a way that is convenient for users, it is essential to have technology that can thoroughly detect subtle changes in people’s speech.

In particular, “speech” can be an important “signal” to assess people’s health conditions. By analyzing speech well, it is possible to diagnose dementia early in the elderly and detect depression at an early stage. He explained, “All the brain’s nerve cells are connected. When a person suddenly experiences severe stress or dementia appears, there is a flow that seems like cells are intermittently disconnected.” He added, “(Simply put) when the flow of cells appears to be interrupted, the early symptoms manifest in ‘speech.’” This means that difficulty in quickly retrieving memories affects the speed of speech expression, which is expressed as a mental health risk signal.

Combining this idea and technology, the ambitious service launched after the company’s founding in 2019 is called “Malgeunnae Chingu” (Clear Voice Friend). He said, “Although the app was opened last September, we are starting full-scale promotion from this week after further preparations.” The goal is to gradually expand the service horizon. While advancing the technology through B2C (business-to-consumer) services like Malgeunnae Chingu, the plan is to create synergy effects by providing the speech analysis AI technology that enables the service to various companies, local governments, and organizations that need this technology.

- What kind of company is Baikal AI?

▲ Baikal AI is an AI software startup founded in 2019. Currently, the focus is on analyzing Korean speech through natural language processing (technology that uses computers to analyze human language) to provide information about people’s health conditions. Speech is a core means of human communication, and changes occurring in the brain can be detected through speech. While the current service analyzes users’ health conditions, the company plans to expand its technology development to diagnose cognitive impairments through speech data in the future.

- Are there services that diagnose health conditions through speech analysis?

▲ Yes. We provide a service called “Malgeunnae Chingu” through an app. When users input speech into the app, it diagnoses various health conditions and analyzes fluency. Fluency analyzes an individual’s language habits?such as whether they use a diverse vocabulary sufficiently or if they use meaningless repetitive phrases or words?through comprehensive analysis. Dementia diagnosis is a representative example of health condition analysis. The app informs users of their dementia risk level based on the voice they input.

- Is there a particular reason for analyzing an individual’s speech and fluency? Are there special benefits for users?

▲ Of course, it may not be very necessary for younger people. Our service is more useful for relatively older individuals. There is more information in how people speak than one might think; for example, early symptoms of depression can be diagnosed through speech analysis. Most people do not speak perfectly like announcers. However, if speech habits include stuttering or frequent errors, it may indicate some cognitive impairment or severe depression, which can be detected quickly.

- So, should this be understood as a voice version of a chatbot?

▲ It is different from a chatbot. If a chatbot is defined as a kind of voice search service, our AI technology listens carefully to people’s speech and helps improve cognitive abilities by asking follow-up questions.

- What does it mean to improve cognitive abilities?

▲ “Malgeunnae Chingu” does not stop at speech analysis but also provides a kind of training service to enhance users’ cognitive abilities. It offers various reading materials to encourage accurate reading and analyzes users’ speech to ask appropriate questions, enabling continuous communication. Through our service, users can talk about their interests or emotional states. By communicating and repeatedly speaking, it helps improve cognitive abilities. Of course, this service is more suitable for the elderly, who have fewer opportunities to speak regularly, than for younger people. Therefore, most users of our service are in their 40s to 70s. The idea is based on the fact that the more one speaks, the better the cognitive function.

- Then, can Baikal AI be seen as a company with a B2C business model centered on speech analysis services like Malgeunnae Chingu?

▲ We are also conducting B2B business. We are in talks to provide our technology to local governments and large corporations emphasizing ESG management. For example, one local government is preparing a project to call elderly people living alone and check their mental health conditions such as dementia, and they are considering using our technology in this service.

- What other projects are currently being prepared?

▲ We plan to integrate a “Safe Call Service” into the Malgeunnae Chingu service before Parents’ Day. Through our app, users will make calls, and the recorded call audio will be analyzed for speech habits and dementia possibility. This service is currently provided free of charge, but some paid services will be added in the future.