ARAUSv2: An Expanded Dataset and Multimodal Models of Affective Responses to Augmented Urban Soundscapes

The ARAUS (Affective Responses to Augmented Urban Soundscapes) dataset consists of a five-fold cross-validation set and independent test set of subjective perceptual responses to augmented soundscapes presented as audio-visual stimuli. However, key limitations in its original release included a disproportionate number of participants being young university students and a relatively small test set. We aim to address this by publishing ARAUSv2, which adds responses from participants to the cross-validation set from an older, non-student population, as well as responses from additional participants in a substantially larger test set consisting of new urban soundscapes recorded in a variety of settings in Singapore. The additional responses were collected in a similar fashion as the initial release, with participants rating augmented soundscapes (made by digitally adding maskers to urban soundscape recordings) on how pleasant, annoying, eventful, uneventful, vibrant, monotonous, chaotic, calm, and appropriate they were. We also present a sample of multimodal prediction models for the ISO Pleasantness and Eventfulness of the augmented soundscapes in ARAUSv2. The multimodal models use participant-linked information such as demographics and responses to psychological questionnaires, as well as visual information from the stimuli, which the baseline models presented in the initial ARAUS dataset did not utilize.