Understanding sound source localization using manifold learning
The Journal of the Acoustical Society of America · 01.10.2025 · 158 (4_Supplement) · A72-A72
DOI: 10.1121/10.0040167 ↗Kurz gefasst
Fachartikel in der Zeitschrift The Journal of the Acoustical Society of America, erschienen am 01.10.2025. Verfasst von Ahsan J. Cheema.
Abstract (Originalsprache)
Human sound-source localization relies on subtle inter-aural time and level differences encoded in the spatiotemporal firing patterns of auditory nerve fibers integrated by central mechanisms. We developed a quantitative framework to relate low-dimensional embeddings of binaural neurogram data to understand the spatial source localization information carried by the auditory nerve. After processing the common environmental sound stimuli through head related transfer function (HRTF) and simulating the auditory nerve response from both ears, various manifold-learning techniques were applied to represent the high-dimensional neural spiking data on a low-dimensional manifold. Embeddings for each technique were then related to spatial coordinates and spectral properties of sound source. It was found that Uniform Manifold Approximation (UMAP) embeddings preserve the azimuth and elevation information from AN spiking data and learnt manifold showed transition patterns wrt. the spatial coordinates. Further, we also show that after hearing loss, the manifold representation collapses, and the classes of spatial location start overlapping which could help explain sound localization difficulties by individuals after hearing-loss. Using this work, we can design perceptual loss functions to design beamforming and amplification strategies for hearing aids that specifically try to restore binaural manifold representations after hearing loss. [Work supported by NIH grant R01-DC07910.]
Zitieren
APA
Cheema, A. J. (2025). Understanding sound source localization using manifold learning. The Journal of the Acoustical Society of America, 158(4_Supplement), A72-A72. https://doi.org/10.1121/10.0040167
Vancouver
Cheema AJ. Understanding sound source localization using manifold learning. The Journal of the Acoustical Society of America. 2025;158(4_Supplement):A72-A72. doi:10.1121/10.0040167
BibTeX
@article{Cheema2025,
author = {Cheema, Ahsan J.},
title = {Understanding sound source localization using manifold learning},
journal = {The Journal of the Acoustical Society of America},
year = {2025},
volume = {158},
number = {4_Supplement},
pages = {A72--A72},
doi = {10.1121/10.0040167}
}
EUVIMED