Deeply (디플리) secures 3rd straight ICASSP paper acceptance, expanding acoustic AI in manufacturing and public safety
The AMW Read
Confirms growing acoustic AI sub-segment within industrial AI; incremental paper acceptance with production deployments but no funding or valuation disclosed.
Deeply (디플리) secures 3rd straight ICASSP paper acceptance, expanding acoustic AI in manufacturing and public safety
South Korean acoustic AI startup Deeply (디플리) announced that two of its papers were accepted at IEEE ICASSP 2026, marking the third consecutive year its work has been recognized at the world's premier acoustics and signal processing conference. One paper proposes a method for selecting data quality to improve model performance with limited labeled sound data, addressing a bottleneck in industrial deployment. The second introduces a FUN-SSL-based real-time sound source localization architecture using a U-Net structure, balancing accuracy and computational load. Deeply operates the Listen AI solution for quality inspection, safety detection, and predictive maintenance, deployed in domestic and Mexican factories of automotive affiliates and global motor manufacturers, with expansion into public facility safety. The company reports 99.78%+ accuracy over eight years of research and will exhibit at Automate 2026 in Chicago.
This development exemplifies the expanding acoustic AI segment within industrial AI infrastructure, a niche where the data-collection barrier and lack of open-source benchmarks create a moat distinct from image-based quality inspection. As manufacturing digitization accelerates, sound-based anomaly detection offers advantages where visual sensors are insufficient — for example, detecting subtle motor or fastening irregularities in automotive and semiconductor lines. Deeply's steady ICASSP presence and production deployments signal that the technology is moving from academic validation to commercial traction, though the startup faces competition from larger industrial automation players and integrated vision-plus-audio platforms.
The article does not disclose a funding round or valuation, and Deeply's track record is too early to assess cross-substrate impact. However, its focus on real-world industrial data scarcity and compute-efficient inference reflects a pragmatic, segment-specific approach compared to general-purpose foundation model labs. The lack of language dependency in acoustic AI also opens global market access without localization costs, a structural advantage in cross-border industrial SaaS. Analyst take: Deeply is building a defensible position in a vertical with high switching costs — once acoustic models are tuned to specific factory floor environments and labeled datasets are accumulated, replacement barriers rise sharply. The key risk is whether the company can scale beyond automotive and semiconductor verticals into adjacent manufacturing and public safety segments without diluting accuracy or increasing integration complexity.
