RIRMega-Eval

Official benchmark views and evaluation harness artifacts built from:

• mandipgoswami/rirmega :[https://huggingface.co/datasets/mandipgoswami/rirmega], arxiv: [https://arxiv.org/abs/2510.18917]
• mandipgoswami/rir-mega-speech : [https://huggingface.co/datasets/mandipgoswami/rir-mega-speech], arxiv: [https://arxiv.org/abs/2601.19949]

This HF repo stores:

• metadata parquet
• fixed splits
• checksums + manifest
• (optional) a small core audio subset

Tasks

• v1_param_estimation: RIR -> RT60, EDT, DRR, C50, C80, Ts
• v1_auralization_consistency: (dry + RIR) -> convolved comparisons

How to evaluate

Use the official evaluator:

• Code: https://github.com/mandip42/rirmega-eval
• CLI: python scripts/evaluate.py ...

Related benchmark: RIR-Bench-Hard

For a dedicated adversarial evaluation benchmark built in the same research program, see RIR-Bench-Hard — 2,600 synthetic RIRs across five deliberately hard regimes (long-T60 halls, coupled rooms, near-field sources, low-DRR, non-shoebox geometries) with standardized est_hard leaderboard splits, a reproducible evaluation harness, and per-RIR acoustic metadata (T60, DRR, C50, EDT). RIR-Bench-Hard is the stress-test complement to this dataset: use it to measure how acoustic-parameter-estimation and dereverberation methods hold up under extreme conditions.

• Benchmark: https://huggingface.co/datasets/mandipgoswami/RIR-Bench-Hard

Related eval sets

Part of the same room-acoustics research program:

• mandipgoswami/audio-eval-suite — small audio-robustness eval set (clean / reverberant / noisy conditions) with reference RT60-regression and reverb-classification tasks.
• mandipgoswami/reverb-speech-mini — tiny dry/wet reverberant-speech pairs for fast dereverberation / robust-ASR sanity checks.
• mandipgoswami/RIR-Bench-Hard — adversarial room-impulse-response benchmark across five hard regimes.

Citation

See CITATION.cff in the code repo.