Stereo personal sound zone rendering with spatially adaptive neural network and realistic acoustic modeling

A stereo personal sound zone (PSZ) rendering framework employing a spatially adaptive neural network for two head-tracked listeners is proposed. The framework facilitates independent optimization of acoustic fields at each ear of the two listeners. Unlike traditional mono-PSZ approaches, separate filters are generated for the listeners' left and right ears, establishing two distinct bright zones and effectively minimizing acoustic energy in dark zones intended for another listener. To reflect realistic acoustic conditions, the framework relies on including effects of transducer directivity and listener HRTF as well as anechoically measured impulse response of the transducers. Simulation and experimental results demonstrate improved inter-zone isolation (IZI) and inter-program isolation (IPI), and enhanced spatial audio fidelity compared to previous mono implementation. The proposed approach offers precise stereo acoustic field manipulation in realistic scenarios while preserving real-time adaptability to listeners' movements, highlighting its potential for immersive spatial audio applications, and laying the groundwork for future integration with crosstalk cancellation to further enhance spatial realism.