Benchmark explorer
RIR characteristics, from the real dataset
Interactive distributions computed directly from the published RIR-Mega metadata. Only fields that actually carry data are charted — no metadata is invented.
mandipgoswami/rirmega metadata_v2.parquet (first 400 rows) — 400 rows · mandipgoswami/rirmega (revision main)Not populated in this sample: Source–receiver distance, Reverberation time (T20), Reverberation time (T30). These columns exist in the schema but are empty in the sampled rows — recomputing them from the raw waveforms is exactly what Track A does.
Direct-to-reverberant ratio (dB)
Higher = dryer / more direct energy.
- n
- 400
- mean
- -6.04
- p5
- -11.19
- p50
- -5.95
- p95
- -0.33
- std
- 3.64
Clarity C50 (dB)
Early (≤50 ms) vs. late energy; speech clarity.
- n
- 400
- mean
- 9.64
- p5
- 1.74
- p50
- 9.31
- p95
- 18.53
- std
- 5.20
Clarity C80 (dB)
Early (≤80 ms) vs. late energy; music clarity.
- n
- 400
- mean
- 15.51
- p5
- 5.05
- p50
- 14.58
- p95
- 28.44
- std
- 7.19
Definition D50
Early-energy fraction, 0–1.
- n
- 400
- mean
- 0.86
- p5
- 0.60
- p50
- 0.90
- p95
- 0.99
- std
- 0.12
Early decay time (s)
Decay slope over the first 10 dB.
- n
- 400
- mean
- 0.41
- p5
- 0.14
- p50
- 0.34
- p95
- 0.92
- std
- 0.25
DRR vs. C50 (per-RIR)
Each point is one real RIR. Dryer RIRs (higher DRR) tend to show higher clarity — a sanity check that the two published descriptors move together as room acoustics predicts.