AI Paper Atlas

Research map / 3D vision / 3D generation

SoundCam: A Dataset for Finding Humans Using Room Acoustics

Mason Long Wang, Samuel Clarke, Jui-Hsien Wang, Ruohan Gao, Jiajun Wu

NeurIPS 2023 · 3D vision › 3D generation · approach: Visual-acoustic scene modeling · 4 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use multichannel VGGish trained from scratch instead of pretrained VGGishVGGish · Treated Room sine RIR, 10 micslocalization error (cm): 93.6 (53.1) -> 18.1 (13.1)Improved
Reduce microphones from 10 to 1 (ablation)VGGish (multichannel) · Treated Room sine RIRlocalization error (cm): 18.1 -> 71.6 (50.4)Hurt
Reduce microphones from 10 to 4 (ablation)VGGish (multichannel) · Treated Room sine RIRlocalization error (cm): 18.1 -> 17.2 (16.2)No clear effect
Use music-derived RIR instead of sine-sweep RIRVGGish (multichannel) · Treated Roomlocalization error (cm): 18.1 -> 31.4 (23.9)Hurt
Test on a different human than trained onVGGish (multichannel) · Living Room music RIRlocalization error (cm): 25.6 (18.4) -> 50.7 (39.7)Hurt
Train on base layout, test with fabric panelsVGGish (multichannel) · Treated Roomlocalization error (cm): 18.1 -> 132.6 (107.8)Hurt
Increase training data from 25 to 800 points (ablation)VGGish (multichannel) · Living Roomlocalization error (cm): 166.5 (61.2) -> 27.9 (22.0)Improved
Downsample audio from 48 kHz to 16 kHz (ablation)VGGish (multichannel) · Treated Room localization, 4 micslocalization error (cm): 17.2 (16.2) -> 21.8 (16.1)Hurt
Downsample audio from 48 kHz to 16 kHz (ablation)VGGish (multichannel) · Treated Room localization, 1 miclocalization error (cm): 71.6 (50.4) -> 49.4 (54.3)Improved
Downsample audio from 48 kHz to 16 kHz (ablation)VGGish (multichannel) · Treated Room human identification, 10 micsclassification accuracy (%): 82 -> 63Hurt
Use silence instead of RIR for identificationVGGish (multichannel) · Treated Room identification, 10 micsclassification accuracy (%): 82 -> 20Hurt
Use raw music instead of music-derived RIR for detectionVGGish (multichannel) · Treated Room detection, 10 micsclassification accuracy (%): 99 -> 75Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy