Multi-Language Intelligent Voice Interaction Under Noise: How the Noise Type and Signal-to-Noise Ratio Shape Comprehension Performance, Pupil Activity and User Experience Quality

Intelligent voice interaction is increasingly used in noisy daily environments, yet the combined effects of noise type, signal-to-noise ratio (SNR) in native and second-language context remain unclear. We designed eye movement experiments in which participants interacted with an intelligent voice system under various noise conditions in native (Study 1) and second-language environments (Study 2). Additionally, data were scraped from social media (Study3). The findings suggest that (1) the effects of noise type, the SNR, and their interaction vary significantly in terms of comprehension performance and user experience in multi-language intelligent voice interaction; (2) some moderate, rhythm-based or familiar noise improves native-language intelligibility and reduces pupil-indexed listening effort, revealing a “beneficial-noise window”; and (3) the “beneficial-noise window” disappears in second-language use and in real-world posts. This research provides specific actionable practical implications for improving the design of voice interaction systems to accommodate various noise environments.