Search Papers | Poster Sessions | All Posters
Poster A79 in Poster Session A - Tuesday, August 6, 2024, 4:15 – 6:15 pm, Johnson Ice Rink
A Computational Framework for Sound Localization in Auditory Scenes
Lakshmi Narasimhan Govindarajan1,2,3, Ajani Stewart1,2, Sagarika Alavilli1,4, Josh H. McDermott1,2,3,4; 1Department of Brain and Cognitive Sciences, MIT, 2McGovern Institute for Brain Research, MIT, 3K. Lisa Yang Integrative Computational Neuroscience (ICoN) Center, MIT, 4Speech and Hearing Biosciences and Technology, Harvard
Humans routinely localize sounds in the world, but little is known about localization abilities in the presence of concurrent sources. We developed a model of multi-source localization by training a model to generate a probability distribution over locations given binaural audio input. We conducted an experiment to measure human multi-source localization in scenes composed of multiple natural sounds at different locations in azimuth and elevation. Human localization became less accurate as the number of sources was increased, showing marked impairments even for two sources compared to one. The model replicated this dependence on the number of sources, suggesting that human limitations are likely inevitable consequences of sampling the spatial world with only two sensors.
Keywords: Auditory scenes Sound localization Circular regression