Deeply CEO Suji Lee: “Defects Invisible to the Human Ear Can Be Detected with Sound AI”

Beyond Vision AI

Manufacturing is rapidly automating with the adoption of AI and robots, but the area of judging part defects by sound still depends heavily on human ears. Deeply CEO Suji Lee met this industrial challenge with proprietary Sound AI technology and has become a leading figure in the machine-hearing field.

Lee said that sound-based defect inspection is technically more difficult than vision AI and has a higher barrier to entry. She noted that even global companies such as Porsche and BMW have only recently begun exploring the field at the research level, and that very few companies worldwide have both large-scale data and real mass-production deployment experience.

In the past, factories had to build expensive anechoic chambers or rely on specialized inspectors to identify defective parts. These processes required significant time and labor, reducing productivity. Deeply solved this problem with its Listen AI Industrial solution. Using an in-house AI acoustic analysis model, it precisely analyzes operating and fastening sounds from key parts such as actuators, motors, and gears to determine defects immediately.

Lee said that even if a product appears fine on the outside, subtle sound-based defects are a unique area that vision AI cannot detect. That insight became the foundation for the solution’s development.

Proprietary Sound AI

Deeply’s technology is built on three main strengths. First, it uses a proprietary AI acoustic analysis model trained on manufacturing-specific datasets. Rather than relying on open-source models, the company built its own model by training on hundreds of thousands of real-world factory samples.

Second, the model combines anomaly detection and denoising. AI detects abnormal signals that deviate from normal patterns while effectively removing surrounding noise, improving accuracy. Deeply says its ability to identify a target sound against background noise, measured through signal-to-noise ratio, is a core indicator of technical strength.

ROI Calculator

Discover the potential cost savings of robotic automation over a 20-year system life

This calculator compares your current manual labor costs against the total cost of owning and operating a robotic system over its 20-year lifespan.

EXPLORE TODAY

Listen AI Industrial can complete inspection in under one second per product and achieves an inspection accuracy of 99.78%. The company says this helps client companies cut operating costs by more than 60% and achieve tangible ROI. Lee said that factory noise often reaches around 100 decibels, while defect signals can be as subtle as 1.77 decibels, making Listen AI Industrial a uniquely capable solution for identifying extremely small differences.

Easy Deployment

Another strength of Deeply’s solution is ease of deployment. It can be installed immediately on production lines by connecting acoustic sensors and devices without major equipment construction. It also integrates flexibly with existing systems and can respond quickly when a process changes or a line expands.

The workflow is straightforward. A microphone sensor on the conveyor belt collects acoustic data, the dedicated AI server analyzes it, and the results are sent to the factory management system (MES) or automated control system (PLC). Managers can monitor the process 24 hours a day through a display, helping accelerate the realization of unmanned smart factories.

Lee said that customers who initially adopted the solution for just one process often expand it across the entire production line after seeing the results, from motor parts to finished products. She added that inquiries are now coming not only from Korea but also from overseas.

Deeply currently supplies its solution to many domestic manufacturing sites, including major companies such as Hyosung Electric and Korail, and has also expanded into global markets such as the United States and Thailand.

Lee said machine hearing is more than just a tool for finding defects; it is a core field of Physical AI that is responsible for product quality and corporate trust. She added that Deeply aims to become a specialized company that sets the global standard for acoustic analysis in industrial settings.