Structure from Silence: Learning Scene Structure from Ambient Sound

Ziyang Chen, Xixi Hu, Andrew Owens
Proceedings of the 5th Conference on Robot Learning, PMLR 164:760-772, 2022.

Abstract

From whirling ceiling fans to ticking clocks, the sounds that we hear subtly vary as we move through a scene. We ask whether these ambient sounds convey information about 3D scene structure and, if so, whether they provide a useful learning signal for multimodal models. To study this, we collect a dataset of paired audio and RGB-D recordings from a variety of quiet indoor scenes. We then train models that estimate the distance to nearby walls, given only audio as input. We also use these recordings to learn multimodal representations through self-supervision, by training a network to associate images with their corresponding sounds. These results suggest that ambient sound conveys a surprising amount of information about scene structure, and that it is a useful signal for learning multimodal features.

Cite this Paper

BibTeX
@InProceedings{pmlr-v164-chen22b, title = {Structure from Silence: Learning Scene Structure from Ambient Sound}, author = {Chen, Ziyang and Hu, Xixi and Owens, Andrew}, booktitle = {Proceedings of the 5th Conference on Robot Learning}, pages = {760--772}, year = {2022}, editor = {Faust, Aleksandra and Hsu, David and Neumann, Gerhard}, volume = {164}, series = {Proceedings of Machine Learning Research}, month = {08--11 Nov}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v164/chen22b/chen22b.pdf}, url = {https://proceedings.mlr.press/v164/chen22b.html}, abstract = {From whirling ceiling fans to ticking clocks, the sounds that we hear subtly vary as we move through a scene. We ask whether these ambient sounds convey information about 3D scene structure and, if so, whether they provide a useful learning signal for multimodal models. To study this, we collect a dataset of paired audio and RGB-D recordings from a variety of quiet indoor scenes. We then train models that estimate the distance to nearby walls, given only audio as input. We also use these recordings to learn multimodal representations through self-supervision, by training a network to associate images with their corresponding sounds. These results suggest that ambient sound conveys a surprising amount of information about scene structure, and that it is a useful signal for learning multimodal features.} }
Endnote
%0 Conference Paper %T Structure from Silence: Learning Scene Structure from Ambient Sound %A Ziyang Chen %A Xixi Hu %A Andrew Owens %B Proceedings of the 5th Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2022 %E Aleksandra Faust %E David Hsu %E Gerhard Neumann %F pmlr-v164-chen22b %I PMLR %P 760--772 %U https://proceedings.mlr.press/v164/chen22b.html %V 164 %X From whirling ceiling fans to ticking clocks, the sounds that we hear subtly vary as we move through a scene. We ask whether these ambient sounds convey information about 3D scene structure and, if so, whether they provide a useful learning signal for multimodal models. To study this, we collect a dataset of paired audio and RGB-D recordings from a variety of quiet indoor scenes. We then train models that estimate the distance to nearby walls, given only audio as input. We also use these recordings to learn multimodal representations through self-supervision, by training a network to associate images with their corresponding sounds. These results suggest that ambient sound conveys a surprising amount of information about scene structure, and that it is a useful signal for learning multimodal features.
APA
Chen, Z., Hu, X. & Owens, A.. (2022). Structure from Silence: Learning Scene Structure from Ambient Sound. Proceedings of the 5th Conference on Robot Learning, in Proceedings of Machine Learning Research 164:760-772 Available from https://proceedings.mlr.press/v164/chen22b.html.

Related Material