Pre-computed CKA, diffusion-map, and probing results for SE-Probe, the public code release for "Where Does Speech Enhancement Adapt? Probing Study Under Controlled Degradation" (Amar, Ivry, Cohen, 2026).
📄 Paper: arXiv:2512.00482
Layout
snr/cka_snr_.parquet: per-architecture (MUSE, MP-SENet, Demucs) CKA values across additive-noise SNRs and DEMAND noise types, with per-row audio quality metrics (PESQ, STOI, SI-SDR, DNSMOS… See the full description on the dataset page: https://huggingface.co/datasets/yairamr/SE-Probe-data.