docs: survey of recent room-acoustic simulation works and DEISM applicability

[ZeyuXuAudio]

Summary

Adds docs/recent_works_review_for_DEISM.md — a single Markdown report covering:

• High-level summaries of the 5 most recent room-acoustic simulation works/releases (ordered newest first):

  1. RoomDiY (Audio Fusion Bureau, 2026-04-29) — real-time room simulation plug-in with on-device LLM diagnostics
  2. RIR Prediction with Neural Networks — From EDCs to Perceptual Validation (arXiv:2509.24834, 2025-09-30)
  3. DART — Differentiable Acoustic Radiance Transfer (arXiv:2509.15946, 2025-09-19)
  4. DSDN — Differentiable Scattering Delay Networks (DAFx25, 2025-09-02)
  5. GSound-SIR (arXiv:2503.17866, 2025-03-23) — ray-tracing toolkit with high-order Ambisonics

• A one-page summary per work, with LaTeX-flavored equations describing the framework / loss / workflow.

• A three-part applicability analysis against the current DEISM / DEISM-ARG codebase:

  • B.1 Further Acceleration (≈3 pages) — image-source pruning (GSound-SIR pattern), JAX/PyTorch backend (DART pattern), LSTM-EDC surrogate (Lin et al.), FMM-style translation operator factorization.
  • B.2 User Interaction Improvements (≈3 pages) — Parquet/Arrow export of image sources, 3D scene editor, real-time auralization with hybrid early/late rendering, optional LLM "room-doctor", preflight checks.
  • B.3 Framework Extension (≈3 pages) — differentiable DEISM (material/directivity inversion), hybrid DEISM-early + SDN-tail, high-order Ambisonic spatial impulse responses, neural-residual DEISM, non-convex rooms via patch-level ART coupling, public benchmark corpus.

• A phased roadmap connecting each proposal to the work it is inspired by.

Test plan

• N/A — documentation only, no code changes
• Render docs/recent_works_review_for_DEISM.md on GitHub and Read the Docs to verify the LaTeX-style math (uses $…$ and $$…$$) and tables render as intended

https://claude.ai/code/session_012mpgts8f6sNKwHhfunBR7f

---

Generated by Claude Code