A deterministic-stochastic hybrid sound synthesis method for enhancing naturalness of active sound design in electric vehicle

Abstract

To mitigate the lack of driving feedback resulting from the high quietness of electric vehicle cabins, the Active Sound Design (ASD) technology can be employed. However, the sound effects synthesized by mainstream ASD algorithms are often relatively monotonous and exhibit a distinctly artificial quality in actual listening tests. To address this limitation within the context of in-cabin ASD for electric vehicles, a synthesis method based on a Deterministic-Stochastic Hybrid Model (D-SHM) is proposed in this paper. A multi-physics coupling module comprising deterministic and stochastic units is constructed to synthesize acceleration sounds, aiming to enhance sound naturalness. Furthermore, an offline sound synthesis software tool was developed utilizing MATLAB App Designer. This tool maps the model onto a visual operational logic, enabling users to freely combine functional modules according to their requirements and to independently and precisely adjust multidimensional parameters for target sound synthesis. Finally, to validate the naturalness and rationality of the synthesized sounds, a comprehensive evaluation metric for synthesized sound naturalness is constructed to perform an objective quantitative analysis. Various sound samples are generated by systematically adjusting the module weights and related parameters, and evaluation experiments are conducted. The results indicate that variations in different modules demonstrate regular influences on specific secondary metrics. By constraining the synthesis parameters within an optimized range, an excellence rate of 61.1% or higher is achieved for the sound samples. The method proposed in this paper provides a clear basis for parameter adjustment direction and holds practical guiding significance for sound design and optimization.