Data-Driven Optimization Strategy of Microphone Array Configurations in Vehicle Environments

Abstract

Microphone array (MA) speech enhancement is a crucial component of vehicle intelligence. However, the complex acoustic environments and the spatial constraints of array layouts present challenges for the design and implementation of MAs in intelligent vehicles. This study proposes a data-driven optimization strategy for constructing the optimal MA configuration in-vehicle environments. We first developed a novel in-vehicle noise model that considers azimuth and elevation angles by defining a search region for microphone elements in a plane. Subsequently, based on the in-vehicle noise model, we conducted sound field modeling to ensure the designed MA is compatible with the complex acoustic environments inside vehicles. Utilizing this sound field model, we formulated a specialized optimization algorithm to devise the optimal configuration of the MA. Finally, the designed array configuration was constructed using an MEMS MA acquisition system, and the array performance was evaluated in real driving environments. Compared to conventional MA configurations, comprehensive experiments indicate that the designed MA enhances performance by increasing the short-time objective intelligibility (STOI) scores by 13.9%, improving the output signal-to-noise ratio (SNR) levels by 53.3%, and ensuring robustness in complex in-vehicle acoustic environments.