A subjective evaluation prediction method for intelligent cockpit sound field zoning based on RTA neural network

Addressing the lack of subjective perception evaluation for sound field zoning within intelligent cockpits, the novel evaluation method based on the Residual-Transformer-Attention (RTA) neural network is proposed. Objective acoustic metrics and subjective listening scores were collected from in-vehicle sound field zoning experiments. The comprehensive zone comfort was developed as a metric from subjective evaluations. The RTA neural network introduces Transformer encoder and attention pooling mechanism to enhance the global feature modeling ability, and combines cost-aware loss function to improve the model’s learning ability for high error samples. The results show that the proposed RTA model achieves higher prediction accuracy, exceeding 97.5%. It can effectively establish a nonlinear mapping relationship between objective sound field metrics and subjective evaluations. The methodology presented in this paper offers an effective approach to reduce the cost of subjective listening experiments and improve the efficiency of design and optimization of intelligent algorithms for sound field zoning.