engleski

Auditory attention modeling within SONORUS ESR 10

This contribution presents the outcomes of the research conducted within the scope of the SONORUS ESR 10 project theme. Within this theme, the influence of human auditory attention on the analysis and design of urban soundscapes is investigated. Several important attention mechanisms are evaluated: saliency, object formation and personal beliefs. Firstly, an auditory saliency model inspired by neural processing was created using features based on spectro-temporal modulations. Testing the model with environmental sounds that include salient events demonstrates its ability to correctly predict the saliency of the sound. Furthermore, object formation and resulting sound recognition was improved using an already developed computational auditory attention model based on a recurrent neural network. The output of the trained model shows that the sounds to which the model attends correspond to the reported sounds from the visitors of urban parks. Finally, to test if people’s beliefs influence their attention, their viewpoint on tranquility was evaluated. As a result, it is determined that the visitors of the investigated parks who belong to groups that associate tranquility to natural sound sources and silence hear more mechanical sounds.Future urban sound planners could use all proposed models as tools for steering their decisions in evaluating human attention and perception of sounds in present or planned environments.

human attention, artificial models, tranquil areas

nije evidentirano