Deep Learning-Based No-Reference Video Streaming QoE Estimation Using WebRTC Statistics

Web Real-Time Communication (WebRTC) enables low-latency communication and is commonly used for realtime video streaming. To ensure acceptable Quality of Experience (QoE) for video streaming services, a realtime QoE monitoring model is required. While Full-Reference (FR) models have a high correlation with subjective evaluations, they are not suitable for realtime monitoring. In contrast, No-reference (NR) models exhibit realtime performance but lower correlation. In this paper, we propose a deep learning-based NR model that achieves the same level of QoE estimation accuracy as the FR model. For the training data, we used the QoE computed from Video Multimethod Assessment Fusion (VMAF), an FR model known to be highly correlated with subjective ratings. The proposed model is trained to estimate QoE based on WebRTC statistics. Our experiments show that the proposed NR model outperforms the existing model in terms of Root Mean Squared Error (RMSE) and Pearson Correlation Coefficient (PCC).