Medisensing to Expand Medical and Daily Life Markets with AI That Understands Situations Through Sound

- Medisensing, founded by Professor Seong-eun Kim of Seoul National University of Science and Technology, is developing Sound State Intelligence (SSI) technology that interprets the context of non-verbal sounds like screams, baby cries, and respiratory distress.
- Unlike general-purpose models, Medisensing’s AI uses contextual reasoningto analyze frequency, intensity, and duration relative to a home’s normal noise environment, providing actionable data for elderly care, childcare, and public safety.
- The company has successfully developed a lightweight Edge SDKthat functions on devices like smartphones without server connections and is currently preparing for seed investment and TIPSmatching in early 2026 to standardize "hearing intelligence" globally.


Social concerns regarding caregiving and safety are steadily increasing. We have reached a point where there is a growing need for technology that can detect moments such as an elderly person living alone collapsing, a baby crying, or someone calling for help in an alleyway. Particularly as non-face-to-face medical treatment and remote care services spread, attempts are being made across various fields to supplement situations where humans cannot directly see or hear through technology.

Seong-eun Kim , CEO of Medisensing / source=IT DongA

The startup Medisensing is responding to these social demands. Specifically, it is venturing into areas ranging from healthcare to safety and caregiving in daily life using Artificial Intelligence (AI) technology that listens to sounds and interprets their meaning and context. Beyond simple voice recognition, the company is developing sophisticated AI technology that understands non-verbal sounds such as screams, cries, and requests for help to judge the situation. We met with Seong-eun Kim, the CEO of Medisensing, to hear about the motivation, strategy, and vision behind the business.

Deciding to Start a Business While Researching the AI Field

Founded in 2024, the name Medisensing combines "Medical" and "Sensing." Kim explained, "I established Medisensing to create technology that can substantially contribute to society by detecting signals generated from the human body and daily life and understanding their meanings."

Kim is also a researcher and a professor in the Department of Artificial Intelligence at Seoul National University of Science and Technology who has published more than 12 papers in top-tier international journals (within the top 10%) over the past five years in the field of AI for understanding biosignals, brainwaves, and sound signals. While conducting research in the AI field, he decided to start a business to ensure that the technology does not merely remain at the paper level but actually contributes to society.

Kim started a business based on AI research. / source=IT DongA

"Through joint research with hospitals, I applied AI technology and found that it could identify abnormal respiratory sounds with higher accuracy than expected. Beyond the technical achievement at the time, I clearly saw the possibility that 'this technology could actually help someone.' I thought that if this technology could be implemented using only a smartphone microphone, it could provide basic stethoscopic assistance even in environments with low medical accessibility," recalled Seong-eun Kim. He added, "Through these experiences, I realized the importance of realizing research results as practical value rather than letting them remain in papers. That decision led to the founding of the company. I also wanted my students to directly experience the entire process of technology being materialized into actual products and services beyond the laboratory."

Expanding into a Sound Sensing AI Platform Integrating Core Sounds

The core technology of Medisensing is an AI that accurately recognizes meaningful individual sounds and understands the context even in the complex noise environments of reality. They are developing SSI(Sound State Intelligence) technology, which analyzes sound data to grasp the status and induce action. While general-purpose models have strengths in classification—labeling something as 'this is a cry' based on learned criteria—they inevitably have many errors. In the case of SSI, it uses the normal sound environment of each household as a reference to judge 'how meaningful this sound is compared to the usual.' It essentially answers the 'why' and 'what' of the sound by combining time-series patterns with LLM-based contextual reasoning. For example, it interprets a child's continuous high-pitched crying not just as a simple cry, but as pain or a risk of abnormal signs.

Kim stated, "When a general-purpose model hears a baby's cry, it usually outputs a probability like 'baby cry 0.72.' From the user's perspective, it is difficult to connect that to 'so what should I do now.' SSIorganizes the information into a form that the user can immediately judge, such as when the crying started, how long it lasted, the intensity, and whether it recurred within the last 10 minutes." He continued, "Even for the same baby's cry, the volume of the TV differs in every house, and every child has a different usual crying pattern. SSI first learns the normal sounds and patterns of that specific house and child, and organizes it as an event when a change deviating from that standard is detected."

Based on this, Medisensing developed a beta version of iMedic, a self-stethoscopic tool app for AI-based pediatric respiratory evaluation, and introduced it in 2025. The core of iMedic is to measure a child's respiratory sounds with a smartphone microphone to determine abnormalities and then record and manage them. The focus was on making it easy for anyone to operate at home and allowing the sound to be transmitted to medical staff if necessary.

However, it was confirmed that additional procedures, such as medical device certification, were required for the official launch of iMedic. Consequently, the beta service was terminated. Seong-eun Kim said, "The core technology of iMedic was based on research results already published in academic papers. Although it did not lead to commercialization, the experience of implementing technology that remained in papers into an actual MVP(Minimum Viable Product) form and verifying it in the field was very meaningful."

Beta version of iMedic developed by Medisensing / source=IT DongA

Medisensing is also focusing on developing technology to recognize five to six core individual sounds, such as baby cries and screams. Kim explained, "We have developed a model that can stably recognize baby cries even in environments where TV sounds, daily noise, and surrounding conversations are mixed. We are expanding this technology into a sound sensing AI platform that integrates various core sounds rather than being limited to specific sounds."

Above all, Medisensing seeks differentiation by lightening its core technology so that it can operate without large-scale server resources. Medisensing provides this technology in the form of an Edge SDK. They plan to increase utility by designing it so that it can be installed and used on any device connected to a microphone, such as a smartphone.

Accordingly, the technology utilization scenarios are diverse. Representative examples include safety systems installed in alleyways or public spaces to detect screams or requests for help at night, as well as caregiving assistance technology in homes that recognizes everything from a baby's cry to whimpering.

"What Medisensing wants to create is not just a simple sound recognition function. It is hearing intelligence that assists moments that humans had to listen to and judge directly. Although we started in medicine, we are expanding the technology in a direction that helps human safety and care in daily life and society as a whole," said Kim.

Solving Difficulties in Data Acquisition and Real Noise Environments by Changing the Approach

The journey to developing the technology for Medisensing was not easy. The biggest difficulty was undoubtedly data acquisition. Kim confessed, "Unlike images or text, public data for sound is very limited. In particular, sound data that includes the context of actual situations hardly exists. In the case of medical sounds, practical constraints on development speed were added because clinical validation and ethical considerations are essential."

Another problem was that sound recognition technology almost always has to operate 'within noise' in actual usage environments. Kim said, "Models that worked well in a laboratory environment often saw a sharp decline in performance in real environments where daily noises like TV and conversation were mixed. However, it was not easy to secure and reproduce such realistic environmental noise itself."

That was not the end. There was also a large gap between research and business. Kim stated, "While research requires sufficient verification and repetition, business requires rapid execution and pivoting. Satisfying these two demands simultaneously was a constant concern during the technology development process."

Medisensing has changed its approach to address a series of problems / source=IT DongA

Nevertheless, they did not give up. They changed their approach to solve the problem. Instead of trying to understand all sounds comprehensively from the beginning, they established a strategy to define and recognize sounds one by one, starting with those that have clear meanings and high social utility. As a result, they are currently focusing on developing technology that selects five to six core individual sounds, such as baby cries, screams, and requests for help, and defines and recognizes each as an independent unit of meaning.

"After changing the strategy, we were able to clearly limit the scope of the problem and rapidly increase the technical maturity. Also, to create conditions similar to actual environments, we introduced a method of designing environmental noise into various scenarios and reproducing them by combining them with Generative AI. For example, by artificially generating noise conditions assuming actual usage environments such as homes, alleys, and indoor public spaces, and utilizing them for learning and verification, we were able to develop a model strong against noise environments. We lightened the model so that it works properly even in real environments and designed it to operate stably with very few computational resources," explained Kim.

As a result, through a series of strategies, Medisensing was able to secure a lightweight sound recognition AI module that does not miss meaningful sounds even in realistic noise environments. This has become the core foundation of the sound sensing AI platform that Medisensing is currently developing.

The biggest challenge for Medisensing currently is to accurately identify what 'sound-based technology' the market needs right now. Kim said, "Technically, we have secured a significant portion of AI technology that recognizes individual sounds and operates stably in noisy environments, but we are at a stage of clearly defining what problem this technology solves to become an 'absolutely necessary technology.'"

Parallel to technology development, Medisensing is making efforts to verify actual needs through interviews and PoC(Proof of Concept) with various industry stakeholders. They are confirming one by one which sound recognition functions are most desperate in the field, what form they should be provided in to have value as a service, and what business model should be created based on this.

Aiming for Step-by-Step Growth Through Parallel Technology and Market Validation
The strength of Medisensing is that it is not based on an idea created in a short period, but on research capabilities and achievements accumulated in the laboratory over a long time. Top-tier international journal papers and a patent portfolio are also important assets that are difficult to possess in the early startup stage. In addition, the technical capability to simultaneously develop and distribute Cloud API and Edge SDK is considered a factor that increases the possibility of collaboration with various industrial partners.

Medisensing participated in Next Stage: Global IR for Growth-Stage Startups / source=IT DongA

Based on this business feasibility, Medisensing was selected for the Initial Startup Package in the deep-tech field by the Seoultech Startup Support Foundation in 2025. Kim expressed satisfaction, saying, "Through this project, we received initial funding to stably secure labor costs for about five months and established an environment where we could focus on technology development. Not only the financial support, but the VC meetups, IR material preparation and presentation opportunities, and networking programs were of great practical help. These processes went beyond simple support and helped Medisensing systematically prepare to move to the next stage."

Furthermore, Medisensing has currently completed an initial small-scale investment contract with Seoultech Holdings. They are also preparing to attract seed investment with the goal of linking with TIPS in the first half of 2026. Kim emphasized, "Once stable funds are secured through investment, we plan to expand core technical personnel who can advance sound recognition and situational understanding technology, as well as product planning personnel, to increase the completeness of products and services applicable to the actual market. Rather than short-term expansion, Medisensing's goal is to grow step-by-step while performing technology and market validation in parallel."

According to Medisensing, the mid-to-long-term goal is to build various sound databases and secure hearing intelligence technology that can understand multiple sounds simultaneously and perceive situations. Medicine was an important starting point, and the strategy is to grow it into a core technology that can be expanded to daily life and the industry as a whole in the future.

The vision of Medisensing is "Defining the Standard for Hearing Intelligence." Kim stated, "Humans can perceive the general situation just by hearing sound, but current AI technology is excellent at voice recognition but still has limits in understanding the situational meanings contained in non-verbal sounds. Our goal is to create a common standard and structure for which sounds should be interpreted in what context and with what meaning." He predicted, "Once this standard is established, it can be utilized as a basic infrastructure for AI that understands sound in various industries such as robotics, smart homes, autonomous driving, and caregiving services."

Attention is focused on what kind of results Kim's challenge to connect long-term research achievements to social value will lead to in the market.

By Kui-im Park (luckyim@itdonga.com)