摘要:

Sound field reproduction with undistorted sound quality and precise spatial localization is desirable for automotiveaudio systems. However, the complexity of the automotive cabin acoustic environment often necessitates a trade-offbetween sound quality and spatial accuracy. To overcome this limitation, we propose Spatial Power Map Net, alearning-based sound field reproduction method that improves both sound quality and spatial localization in complexenvironments. We introduce a spatial power map constraint, which characterizes the angular energy distribution ofthe reproduced field using beamforming. This constraint guides energy toward the intended direction to enhance spatiallocalization, and is integrated into a multi-channel equalization framework to also improve sound quality underreverberant conditions. To address the resulting non-convexity, deep optimization that uses neural networks to solveoptimization problems is employed for filter design. Both in situ objective and subjective evaluations confirm thatour method enhances sound quality and improves spatial localization within the automotive cabin. Furthermore, weanalyze the influence of different audio materials and the arrival angles of the virtual sound source in the reproducedsound field, investigating the potential underlying factors affecting these results.