Researchers have released a paper on arXiv proposing a framework that reuses multimodal agents to evaluate sound designs in games. The approach leverages both multimodal learning and reinforcement learning, according to the abstract. The authors note that sound design has a significant influence on gameplay and accessibility, and position their work as an efficient evaluation method.
The framework is aimed at reducing the cost of assessing audio choices in games, though implementation details are not included in the abstract. The paper's focus on objective evaluation suggests that sound design testing could become more systematic, which may be especially relevant for accessibility research.