Modeling outdoor environmental sound levels is a challenging problem. This paper reports on a validation study of two continental-scale machine learning models using geospatial layers as inputs and the summer daytime A-weighted L50 as a validation metric. The first model was developed by the National Park Service while the second was developed by the present authors. Validation errors greater than 20 dBA are observed. Large errors are attributed to limited acoustic training data. Validation environments are geospatially dissimilar to training sites, requiring models to extrapolate beyond their training sets. Results motivate further work in optimal data collection and uncertainty quantification.
This paper describes using both supervised and unsupervised machine learning (ML) methods to improve automatic classification of crowd responses to events at collegiate basketball games. This work builds on recent investigations by the research team where the two ML approaches were treated separately. In one case, crowd response events (cheers, applause, etc.) were manually labeled, and then, a subset of the labeled events were used as a training set for supervised-ML event classification. In the other, (unsupervised) k-means clustering was used to divide a game’s one-twelfth octave spectrogram into six distinct clusters. A comparison of the two approaches shows that the manually labeled crowd responses are grouped into only one or two of the six unsupervised clusters. This paper describes how the supervised ML labels guide improvements to the k-means clustering analysis, such as determining which additional audio features are required as inputs and how both approaches can be used in tandem to improve automated classification of crowd noise at basketball games.
Humans and nature form an intricately coupled system with the ambient soundscape: anthropogenic, biological, and geophysical sources produce the sounds that comprise the soundscape, and, in turn, the ambient sound level affects the behavior and well-being of humans and animals. To assess the impact of the soundscape on both humans and animals, it is necessary to understand how the ambient sound level varies in space and time. A model for the ambient sound level was developed based on an ensemble of machine learning algorithms, which were trained using more than one million hours of ambient acoustic measurements acquired at hundreds of geospatially diverse locations across the United States. The resulting model predicts the ambient sound level based on geospatial features such as nighttime lights, land cover, population, climate, topography, hydrology, and transportation. A database of geospatial features with worldwide coverage was created, and the model was applied to predict the time-varying ambient sound level across the entirety of Earth’s land surface. Furthermore, the relative contributions of anthropogenic and natural sources to the soundscape were estimated by artificially changing the values of various geospatial features and reapplying the model. [Work funded by an Army SBIR.]
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.