Sounds Recognition with High-Performance AI Real-time Processing

SHARP addresses a critical challenge in modern urban lighting: while current EU regulations increasingly require adaptive lighting systems, training the AI models needed to enable them requires HPC resources that are often beyond the reach of SMEs.

TRAILSLIGHT S.R.L., an Italian innovator and developer of the patented TRAILS Light Control System—an acoustic-sensing street lighting platform—has partnered with HPC expert Bi-Rex to overcome this barrier. Using supercomputing infrastructure, SHARP trains a 50-million-parameter foundation model on up to 50,000 hours of urban soundscape data, including synthetic datasets representing rare events such as traffic accidents.

Advanced model compression techniques then reduce this “acoustic brain” to less than 4 MB, enabling real-time inference directly on lamp posts while ensuring full GDPR compliance. The result is an agentic AI network that continuously improves through HPC-enabled cloud updates, with the potential to reduce energy consumption by 40%, light pollution by 60%, and emergency response times by 30%.

Organisations involved: 

End-User: TRAILSLIGHT S.R.L.

Technology expert: Bi-REX