Generative AI-Driven AVAS Sound Design for Enhanced Pedestrian Safety and Sound Quality in Electric Vehicles

This paper explores generative AI methodologies to create sophisticated Acoustic Vehicle Alerting Systems (AVAS) for electric vehicles, aiming to enhance pedestrian safety and improve urban acoustic environments. Leveraging models within the generative AI subset of machine learning, such as GANs (Generative Adversarial Networks) and VAE (Varia tional Autoencoders), the research targets flexible sound design based on specific attributes. A curated database of authentic and synthetic audio samples enables the training of AI models to produce contextually relevant, parameterized sound profiles. These profiles are defined by characteristics like "sporty," "pleasant," or "highly alerting" and meet regulatory requirements while accommodating psychoacoustic preferences. Preliminary results showcase the ability of a prototype AVAS framework to adjust sound frequencies, modulation patterns, and tonal variations, enhancing pedestrian awareness in low-speed scenarios. By integrating these AI-generated sounds into an adaptable audio framework, the research establishes a novel, scalable tool that not only strengthens pedestrian safety but also elevates brand-specific sound quality in urban EV applications. This foundation supports future collaborations to integrate AVAS advancements into series production and extend generative AI applications across automotive sound design.