Urban unmanned aerial vehicle (UAV) operations are increasingly considered for delivery, inspection, emergency response, and urban mobility applications. However, UAV operations in dense urban environments may introduce significant noise exposure to residents and potential ground risk to people and infrastructure. Traditional UAV path planning methods often focus on distance minimization, obstacle avoidance, or energy efficiency, while insufficiently considering cruise-mode noise directivity, population exposure, and ground-risk distribution. This thesis develops a fixed-altitude two-dimensional UAV path planning framework for urban environments that couples cruise-mode noise modeling, learned environmental modeling, learned global planning, and reinforcement-learning-based local control. First, the UAV acoustic model is modified for cruise-mode operation by using directional weighting and Fibonacci hemisphere sampling to improve ray distribution uniformity, and a convolutional surrogate model (U-Net B) is trained to predict an urban sound map directly from the building layout, replacing repeated physical acoustic simulation with a single forward inference.