Research map / 3D vision / 3D generation
SoundSpaces 2.0: A Simulation Platform for Visual-Acoustic Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 changed | Where | Measured | Effect |
|---|---|---|---|
| Use high-speed rather than high-quality rendering | SoundSpaces 2.0 · RIR trajectories, 20 Matterport3D environments | 5 Threads (FPS): 4.0 { $\pm$ 0.1} -> 33.5 { $\pm$ 0.4} | Helped |
| Use high-speed rather than high-quality rendering | SoundSpaces 2.0 · RIR trajectories, 20 Matterport3D environments | Relative RT60 Error (%): 0.0 { $\pm$ 0.0} -> 9.5 { $\pm$ 0.2} | Hurt |
| Add diffraction and correct the indirect-sound pressure bias | SoundSpaces · Replica apartment real RIRs | Mean DRR error (dB): 11.0 dB -> 0.98 dB | Helped |
| Train with continuous sound rather than the discrete SoundSpaces sound setup (ablation) | SoundSpaces-trained DAV-Nav · AudioGoal, continuous space | Success (%): 64.2 { $\pm$ 0.8} -> 0.9 { $\pm$ 0.2} | Hurt |
| Train using continuous spatial and acoustic observations in SoundSpaces 2.0 | DAV-Nav · AudioGoal, continuous space & continuous sound | SPL (%): 0.3 { $\pm$ 0.1} -> 49.3 { $\pm$ 3.0} | Helped |
| Finetune on synthetic SoundSpaces 2.0 impulse responses | SpeechBrain pretrained transformer · LibriSpeech speech augmented with real RIRs | Word Error Rate (%): 29.10 -> 12.48 | Helped |
| Use SoundSpaces 2.0 rather than Pyroomacoustics RIRs for finetuning | SpeechBrain pretrained transformer · LibriSpeech speech augmented with real RIRs | Word Error Rate (%): 16.24 -> 12.48 | Helped |
| Add category- and instance-level acoustic material randomization (ablation) | SoundSpaces 2.0 ASR finetuning · Far-field ASR test with real RIRs | Word Error Rate (%): 12.48% -> 12.04% | Helped |
| Randomize material coefficients uniformly between 0 and 1 (ablation) | SoundSpaces 2.0 ASR finetuning · Far-field ASR test with real RIRs | Word Error Rate (%): 12.48% -> 12.58% | Hurt |
| Test a navigation policy trained in high-speed mode using high-quality rendering | SoundSpaces 2.0 DAV-Nav · AudioGoal navigation | Performance difference: smaller than 1% | No clear effect |
Closest papers by meaning
- Audible World Models: Spatially Aware Sound Generation for 3D Worlds (NeurIPS 2026 · similarity 0.89)
- SoundVista: Novel-View Ambient Sound Synthesis via Visual-Acoustic Binding (CVPR 2025 · similarity 0.88)
- Hearing Anything Anywhere (CVPR 2024 · similarity 0.88)
- Towards Audio-Visual Navigation in Noisy Environments: A Large-Scale Benchmark Dataset and an Architecture Considering Multiple Sound-Sources (AAAI 2025 · similarity 0.88)
- AV-GS: Learning Material and Geometry Aware Priors for Novel View Acoustic Synthesis (NeurIPS 2024 · similarity 0.88)
- OWL : Geometry-Aware Spatial Reasoning for Audio Large Language Models (ICLR 2026 · similarity 0.87)
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.