Deeplee is evolving into a company that analyzes sounds in manufacturing environments using artificial intelligence (AI) to manage quality and safety. Initially focused on baby cries, the company is now expanding into the industrial sound AI market by analyzing machine noises and fastening sounds in manufacturing, including automotive parts.
In an interview on September 30 at Deeplee's office in Mapo, Seoul, CEO Lee Soo-ji described the company as one that supplies auditory modules to automated factories. He stated, "As robots evolve to resemble humans, relying solely on vision will be insufficient, and auditory modules will become essential."
Lee's choice to pursue sound AI as a business stemmed from his background in signal processing. After graduating from KAIST with a degree in electronics, he specialized in brainwave signal processing at Seoul National University. He explained that while brainwaves and sounds may appear to be different data types, they are technically similar. "Broadly speaking, both sound and brainwaves fall under the field of signal processing," he noted. With advancements in frequency division and preprocessing technologies, combined with recent developments in AI and deep learning, it has become possible to analyze more signals at higher resolutions.
Lee likened the advancement of sound analysis technology to the evolution of cameras. Just as blurry images have improved to high resolution, allowing for more information to be captured, sound can now be analyzed in greater detail due to advancements in AI technology.
Deeplee did not initially focus on manufacturing. In 2020, the company launched an application called 'WAAH' that analyzes baby cries. Lee's experience as a parent, struggling to understand what his child wanted based solely on crying, inspired the business idea. "As a parent, I found it difficult to interpret what my child was expressing," he recalled. "I thought that if we gathered and analyzed enough data, it could be effective." While experienced parents can distinguish between hunger and discomfort, first-time parents often struggle, prompting the idea to utilize data.
However, as Deeplee engaged in B2C business, the team faced challenges in determining where to apply their accumulated technology. Ultimately, considering their identity as a technology-based startup and existing technical capabilities, they shifted their focus to B2B. Lee explained, "We are a technology-based startup, and we have accumulated a lot of technology. Both my co-founder and I have engineering backgrounds, so we felt B2B would suit us better."
The differences between B2C and B2B were also evident in their business approaches. Lee noted that in B2C, enhancing user interface (UI) and user experience (UX) for convenience and conducting large-scale online marketing were crucial. In contrast, in manufacturing, it was essential to validate technology on-site and build trust.
Lee remarked, "B2C and B2B feel fundamentally different. In B2C, there were many teams focused on making the consumer experience more convenient through UI and UX, and we conducted large-scale online marketing. However, in manufacturing, it is important to build trust through trade shows and reliable articles."
During the transition, the organization underwent significant downsizing. The team initially expanded to analyze baby cries but reduced its size when that business was discontinued. Eventually, Lee, his co-founder, and two colleagues restarted the B2B business. The team has since grown to 27 employees, with about two-thirds focused on development, working on AI algorithms and product development.
Deeplee identified a limitation in quality inspections that relied on human ears. In automotive parts manufacturing, inspections have traditionally involved listening for sounds indicating whether parts are properly assembled. However, human inspections are difficult to document objectively. The judgment can vary based on the inspector's skill and condition, making it challenging to trace the basis for determining defects. Lee pointed out, "In the existing automotive industry, checks were done by human ears, but now it is difficult to assess objectively using only human ears. We believe we are in the process of transforming these aspects into a system that can document and data-ify them."
Deeplee's industrial sound AI solution, 'Listen AI Industrial,' is utilized for quality inspections of automotive motors, actuators, and bearings, as well as for checking connector and bolt assembly sounds. Currently, the system inspects over 18,000 products per hour in a fully inline manner, with accumulated sound data exceeding 520,000 hours and 20TB.
Lee emphasizes the importance of 'data' in this process. Hundreds to tens of thousands of products are produced daily on manufacturing lines, and the sounds generated during production are recorded along with product information. He explained, "In a mass production environment, some lines produce hundreds of items a day, while others can produce tens of thousands. The sounds produced on those lines are recorded and linked to the specific products, creating data."
This goes beyond merely collecting sounds. It allows for connections between whether a product is normal or defective, the specific product generating the sound, and the production time and process information. This is why Lee identifies the data accumulated on-site as Deeplee's competitive advantage. He stated, "Having such data continuously accumulated is a significant asset."
The application of sound AI in manufacturing is also driven by changes in the automotive industry. As electric vehicles become more prevalent, vehicles are quieter, making quality control of small joints and abnormal sounds increasingly important. The need to differentiate and document subtle differences that were previously checked by human ears has grown.
Lee acknowledges that the sound AI market is still smaller compared to vision AI. However, he believes that as manufacturing environments become more automated, the need for sensors that address not only vision but also other senses will increase. He stated, "Sound AI may be a much smaller field compared to vision or humanoid robots, but I see it as the second sense to vision. If the manufacturing industry continues to move in this direction with automated systems, it can play a significant role across various sensor modalities."
Deeplee aims to target the global market based on its application cases in domestic manufacturing. In 2025, it supplied Listen AI Industrial to H Group's facilities in South Korea and Mexico, and in 2026, it delivered solutions to a global automotive motor manufacturer. Currently, the solution is operational in 30 industrial sites worldwide.
In June, Deeplee secured a new investment of 2.5 billion won from Dev Sisters Ventures, Su Investment Capital, and Nautilus Investment, and in August, it was selected for the TIPS general track by the Ministry of SMEs and Startups.
While the focus has been on enhancing performance tailored to specific production lines, the company plans to improve the overall quality of its products so that more manufacturers can quickly adopt the same products. Lee stated, "While we concentrated on enhancing technology performance for a single line, we now aim to improve the basic product capabilities so that more clients can use the same product. Shortening the implementation and installation times will reduce costs and allow more clients to adopt it."
Expanding into the global manufacturing sector is also a key goal. Lee noted that the company is currently growing at an annual rate of about three times and emphasized the importance of rapidly increasing actual adoption by clients over securing investments or going public.
Lee has a clear vision for Deeplee's future. He aims to create a system where machines can understand not only what humans see but also what they hear. He described Deeplee's direction as a company that creates "the ears of industrial robots." He stated, "While visual modules have dominated automated factories until now, as robots evolve to resemble humans, I believe auditory modules will also be necessary. I want to be a company that supplies those auditory modules."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.