"AI Skilled Worker" That Detects Defects by Sound Emerges... Receiving Love Calls from Overseas

More diverse corporate information on the startups mentioned in this article can be found at Unicorn Factory's big data platform, Data Lab.

What was the background of a researcher who majored in signal processing analysis at KAIST's Department of Electronics and later analyzed human brainwave signals during their master's program in Brain and Cognitive Science at Seoul National University, only to turn their attention to the noisy mechanical sounds of factories? Lee Su-ji, CEO of Deple, founded the sound AI (artificial intelligence) startup Deple in 2017 and is pioneering the industrial acoustic AI solutions sector.

Lee (CEO) said, "Since both brainwaves and sound are time-series signals, I started the business with the vision of adding meaning to sound." He added, "Initially, we operated in areas such as baby sounds and healthcare audio, but from 2024, we discovered the scalability of our technology in analyzing equipment sounds at manufacturing sites and decided to take on this field."

Dpley's core solution is 'Listen AI,' which detects minute mechanical defects amidst the massive noise inside a factory. Even in environments reaching 100 decibels (dB), it identifies defect signals at a level of just 1.77 dB. Lee (CEO) explained, "Beyond simple frequency analysis, the most critical core technology is using Denoising technology to isolate only the target sound we want while removing the vast amount of surrounding noise." The Listen AI solution boasts performance that determines defects with 99.78% accuracy in under one second using a non-contact method.

Listen AI is completely transforming quality inspections that once relied on skilled workers. In some manufacturing sites, there are auditory seam inspection processes where operators listen directly to machine sounds to judge for anomalies or seams. However, even with the human ear's sophistication, repeating tasks in a noisy environment all day inevitably leads to subjective judgment and fluctuating yield rates. Lee (CEO) stated, "Companies that have adopted the Deeply solution are satisfied with its consistent and objective quality maintenance effects, similar to those of machines, and are increasing the number of processes they apply." This not only reduces labor costs but also drives detailed process improvements through objective log management.

It is also excellently utilized to detect internal defects that are not visible on the surface. For example, in the welding process, it detects irregular sounds instead of the consistent sound that should be produced, identifying in real time defects such as air bubbles forming inside. If air bubbles form, they may lead to fatal cracks later even though they appear fine on the outside. This serves to prevent problems that are difficult to identify through visual inspection beforehand.

Smart factory solutions typically require massive infrastructure construction, but Deplee's on-site implementation process is exceptionally smooth. By installing micro sensors and deep learning PCs at each production stage, these components are integrated into a comprehensive factory-wide monitoring system by connecting them with existing Manufacturing Execution Systems (MES) or PLCs. Lee (CEO) stated that "the integration process is structured like an API, allowing on-site personnel to implement it as a turnkey solution in a short time without complex procedures."

Dipli currently supplies solutions to domestic companies ranging from passenger vehicle affiliates to large-scale infrastructure firms producing semiconductors, LINE systems, gas, and steel. In the global market, Dipli plans to prioritize entry into the United States and Japan, where automation rates are high and labor costs are steep. Interestingly, Korean subsidiaries of global companies have adopted Dipli's solutions, prompting inquiries from their headquarters for adoption.

A Korean subsidiary of a company headquartered in the United States or France is exploring overseas expansion based on inquiries from clients, without separate marketing efforts, after reporting innovation cases (use cases) following the introduction of Deplee's solutions to its headquarters.

In Japan, which is considered a conservative market, the company plans to expand its customer base by showcasing technology reviews and demonstrations from a long-term perspective. Lee (CEO) stated, "Japanese companies tend to conduct lengthy analysis and inspections over extended periods," adding, "We have been steadily conducting technology reviews while focusing on business for several years."

Based on acoustic data accumulated for over 50,000 hours, Deeply is focusing its research capabilities on building a universal sound foundation model. In the future, it plans to expand the "machine hearing" ecosystem beyond manufacturing assembly processes to include welding and cutting machining, as well as aerospace and defense sectors.

Lee (CEO) stated, "While Germany and Japan led in acoustic-based sensor technology in the past, our greatest strength lies in combining this with AI algorithms that can be practically applied on-site," adding, "We aim to become the world's leading acoustic AI company in the manufacturing equipment sector."

[Money Today Startup Media Platform Unicorn Factory]