Research map / Image generation / Fast diffusion sampling
Few-shot Acoustic Synthesis with Multimodal Flow Matching
Amandine Brunetto
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add geometry conditioning to audio-only model (ablation) | FLAC · AcousticRooms unseen rooms | FDG: 0.663 -> 0.305 | Improved |
| add audio context to geometry-only model (ablation) | FLAC · AcousticRooms unseen rooms | T60 error (%): 23.41 -> 9.95 | Improved |
| use fine-tuned DINOv3 ViT-S/16 instead of xRIR ViT (ablation) | FLAC (K=1) · AcousticRooms unseen rooms | T60 error (%): 10.91 -> 9.95 | Improved |
| freeze DINOv3 weights (ablation) | FLAC (K=1) · AcousticRooms unseen rooms | R@5 (%): 18.9 -> 15.40 | Hurt |
| train DINOv3 from scratch instead of fine-tuning (ablation) | FLAC (K=1) · AcousticRooms unseen rooms | R@5 (%): 18.9 -> 5.29 | Hurt |
| replace ResNet-18 with frozen pretrained VAE encoder for acoustic context (ablation) | FLAC (K=1) · AcousticRooms unseen rooms | T60 error (%): 9.95 -> 9.40 | Mixed |
| AdaLN for target pose and time plus cross-attention for context, versus cross-attention only (ablation) | FLAC DiT (K=1) · AcousticRooms unseen rooms | T60 error (%): 15.68 -> 9.95 | Improved |
| AdaLN+CA versus in-context conditioning (ablation) | FLAC DiT (K=1) · AcousticRooms unseen rooms | T60 error (%): 69.68 -> 9.95 | Improved |
| AdaLN+CA versus in-context conditioning (ablation) | FLAC DiT (K=8) · AcousticRooms unseen rooms | T60 error (%): 8.12 -> 8.60 | Hurt |
| use stochastic generation instead of fixed noise (ablation) | FLAC · AcousticRooms unseen rooms | T60 error: +6% T60 with deterministic variant | Improved |
| increase guidance weight or number of inference steps (ablation) | FLAC · AcousticRooms unseen rooms | FDG: 0.305 -> 0.280 (20 steps) | Improved |
| more inference steps (ablation) | FLAC · AcousticRooms unseen rooms | T60 error (%) | No clear effect |
Closest papers by meaning
- Acoustic Volume Rendering for Neural Impulse Response Fields (NeurIPS 2024 · similarity 0.88)
- Hearing Anything Anywhere (CVPR 2024 · similarity 0.88)
- Audible World Models: Spatially Aware Sound Generation for 3D Worlds (NeurIPS 2026 · similarity 0.87)
- AV-GS: Learning Material and Geometry Aware Priors for Novel View Acoustic Synthesis (NeurIPS 2024 · similarity 0.87)
- SoundVista: Novel-View Ambient Sound Synthesis via Visual-Acoustic Binding (CVPR 2025 · similarity 0.87)
- Both Ears Wide Open: Towards Language-Driven Spatial Audio Generation (ICLR 2025 · 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.
