Seeing Sound Dataset v1.0.1

This is dataset contains the synthesized soundscapes and crowdsourced audio annotations that accompany the paper,

M. Cartwright, A. Seals, J. Salamon, A. Williams, S. Mikloska, D. MacConnell, E. Law, J. Bello, and O. Nov. "Seeing sound: Investigating the effects of visualizations and complexity on crowdsourced audio annotations." In Proceedings of the ACM on Human-Computer Interaction, 1(2), 2017. https://doi.org/10.1145/3134664

which investigates the effects of soundscape complexity and sound visualizations on the quality and speed of annotations of sound events (i.e. start time, end time, sound class, and proximity).

In this dataset, we varied the soundscape complexity along two dimensions: maximum polyphony (3 levels) and Gini polyphony (2 levels). Maximum polyphony is the maximum number of sound events that occurred simultaneously in the soundscape. Gini polyphony is a measure of the concentration of sound events. For each of the 6 (3 x 2) combinations of complexity levels, we synthesized 10 soundscapes, each of which was 10 seconds long, for a total of 60 soundscapes. Each soundscape was annotated by 90 participants from Amazon's Mechanical Turk. Of these 90 participants, 30 were aided by waveform visualization, 30 were aided by a spectrogram visualization, and 30 did not have any visualization aid. For more details on how this data was collected, please refer to the paper.