A Review of Soundscape Datasets: Challenges and Prospects for Multimodal Research

The field of soundscape research is receiving increasing attention, particularly in machine learning predictions and multimodal applications. For affective computing, these datasets provide labelled audio-visual environmental stimuli and human perceptual responses, enabling emotion prediction from acoustic environment. However, datasets suitable for such studies remain relatively limited. This review explores existing soundscape datasets and the studies including soundscape data, which include real-world audio recordings and emotional or affective annotations, and in some cases, visual content. Additionally, it explores their potential for advancing multimodal soundscape research. The review identifies key challenges in dataset scale, perception frameworks, collection methods, future multimodal applications and accessibility assessing adherence to the FAIR principles across 20 available datasets. It also highlights the importance of emotional and perceptual annotations in soundscape research, emphasising the need for standardised protocols and diverse participant demographics. By providing a comprehensive overview of current datasets and their limitations, this review adopts a data-centric perspective to guide future research efforts in soundscape analysis, multimodal learning applications, and affective computing.