In this letter, we focus on a quality of experience (QoE) aware resource allocation that considers users' time varying channel status and heterogeneous service requirements in multi-cell OFDMA networks. We propose a QoE-based proportional fair (PF) scheduling considering network-wide users' QoE maximization as well as fairness among users. Our utility function is based on the mean opinion score (MOS) model reflecting the application-specific characteristics. We introduce a Bezier curve based continuously differentiable MOS model instead of using the conventional bounded logarithmic MOS model. Then, we propose a utility function, which is proven to be concave, for QoE provision to achieve the global optimality with opportunistic gradient scheduling. Our proposed technique is shown to be easily incorporated with inter-cell interference coordination (ICIC). Simulation results show that the proposed QoE-aware PF scheme significantly improves the network-wide users' average QoE as well as the edge users' performance, e.g., the 5th percentile QoE.