---
license: cc-by-nc-4.0
task_categories:
- audio-to-audio
- audio-classification
tags:
- rir
- acoustics
- benchmark
- evaluation
---

# 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 ...`

## Citation
See `CITATION.cff` in the code repo.