Many studies have documented the relationship between urban morphology and noise exposure, emphasizing the role of urban planning and design in mitigating environmental noise exposure. However, there remains a significant gap in both literature and practice concerning noise mitigation through the optimization of urban design. Addressing this gap, recent advancements in artificial intelligence(AI)- including machine learning, neural network, and evolutionary computation - have been increasingly incorporated into environmental impact assessments and design optimization. In this study, a novel design optimization workflow combines a neural network-based noise prediction model, an urban form generator, and an evolutionary algorithm-based solver has been proposed. This AI-driven approach enables the rapid evaluation of tens of thousands of design alternatives - a scale of analysis impossible with conventional simulation methods due to computational constraints. Tested on a Hong Kong new town site, our approach demonstrated how building form and layout changes led to significant noise exposure alterations. These results highlight AI's potential in early-stage design to optimize building massing and layouts for mitigating noise impact, offering valuable insights for architects and urban planners while dramatically expanding the design solution space beyond what traditional methods could feasibly explore.