This paper proposes a semantic-aware resource allocation (SARA) framework, specifically a flexible duty cycle (DC) coexistence mechanism (SARADC), for 5G-V2X heterogeneous networks (HetNets) using deep reinforcement learning (DRL) and proximal policy optimization (PPO). The framework aims to maximize semantic spectrum efficiency (HSSE) and semantic throughput (ST) in urban environments with high-speed vehicular networking. The authors introduce two metrics: high-speed semantic transmission rate (HSR) and HSSE, and address the coexistence of vehicular and WiFi users in 5G New Radio Unlicensed (NR-U) networks. The proposed SARADC algorithm integrates semantic communication into the communication system, optimizing resource allocation and DC coexistence. Experimental results show that the proposed solution outperforms traditional bit transmission methods in terms of HSSE and ST for both vehicular and WiFi users. The algorithm's effectiveness is demonstrated through simulations, where it achieves faster convergence, higher stability, and superior performance compared to other baseline methods.This paper proposes a semantic-aware resource allocation (SARA) framework, specifically a flexible duty cycle (DC) coexistence mechanism (SARADC), for 5G-V2X heterogeneous networks (HetNets) using deep reinforcement learning (DRL) and proximal policy optimization (PPO). The framework aims to maximize semantic spectrum efficiency (HSSE) and semantic throughput (ST) in urban environments with high-speed vehicular networking. The authors introduce two metrics: high-speed semantic transmission rate (HSR) and HSSE, and address the coexistence of vehicular and WiFi users in 5G New Radio Unlicensed (NR-U) networks. The proposed SARADC algorithm integrates semantic communication into the communication system, optimizing resource allocation and DC coexistence. Experimental results show that the proposed solution outperforms traditional bit transmission methods in terms of HSSE and ST for both vehicular and WiFi users. The algorithm's effectiveness is demonstrated through simulations, where it achieves faster convergence, higher stability, and superior performance compared to other baseline methods.