yairamr commited on
Commit
25877f9
·
verified ·
1 Parent(s): 53c6adb

README: add arXiv link + citation

Browse files
Files changed (1) hide show
  1. README.md +32 -18
README.md CHANGED
@@ -1,26 +1,40 @@
1
- # SE-Probe v0.1.0 — full results parquet snapshot
 
 
 
 
 
 
 
2
 
3
- This directory is the staging area for the full pre-computed CKA / probing
4
- results uploaded to HuggingFace Hub for SE-Probe v0.1.0.
 
 
 
5
 
6
  ## Layout
7
 
8
- - `snr/cka_snr_<model>.parquet` per-architecture (MUSE, MP-SENet, Demucs)
9
- CKA values across additive-noise SNRs and DEMAND noise types, with per-row
10
- audio quality metrics (PESQ, STOI, SI-SDR, DNSMOS family).
11
- - `reverb/cka_reverb_muse_epoch_<N>.parquet` — per-epoch reverb-FT MUSE
12
- probing parquets. Epoch 48 is the final fine-tuned model; the others trace
13
- the fine-tuning trajectory.
14
- - `diffusion_maps/` — pre-computed diffusion-map embeddings per (model, snr,
15
- noise, layer).
16
 
17
- ## Size
18
 
19
- Total ≈ 3.1 GB. HuggingFace Hub stores large parquets via Git-LFS
20
- automatically; no special action required on upload.
 
 
 
 
 
 
 
 
 
21
 
22
- ## Pointer
23
 
24
- Source code that consumes these files lives at
25
- <https://github.com/YairAmar/SE-Probe>. The `scripts/setup.py` helper there
26
- fetches this snapshot when invoked with `--full-data`.
 
1
+ ---
2
+ license: mit
3
+ tags:
4
+ - speech-enhancement
5
+ - interpretability
6
+ - cka
7
+ - probing
8
+ ---
9
 
10
+ # SE-Probe full results dataset
11
+
12
+ Pre-computed CKA, diffusion-map, and probing results for [SE-Probe](https://github.com/YairAmar/SE-Probe), the public code release for *"Where Does Speech Enhancement Adapt? Probing Study Under Controlled Degradation"* (Amar, Ivry, Cohen, 2026).
13
+
14
+ 📄 Paper: <https://arxiv.org/abs/2512.00482>
15
 
16
  ## Layout
17
 
18
+ - `snr/cka_snr_<model>.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 family).
19
+ - `reverb/cka_reverb_muse_epoch_<N>.parquet`: per-epoch reverb-FT MUSE probing parquets. Epoch 48 is the final fine-tuned model; the others trace the fine-tuning trajectory.
20
+ - `diffusion_maps/`: pre-computed diffusion-map embeddings per (model, snr, noise, layer).
21
+
22
+ Total ~3.1 GB. To download into a local SE-Probe checkout: `python scripts/setup.py --full-data`.
 
 
 
23
 
24
+ ## Citation
25
 
26
+ ```bibtex
27
+ @article{amar2026seprobe,
28
+ title = {Where Does Speech Enhancement Adapt? Probing Study Under Controlled Degradation},
29
+ author = {Amar, Yair and Ivry, Amir and Cohen, Israel},
30
+ year = {2026},
31
+ eprint = {2512.00482},
32
+ archivePrefix = {arXiv},
33
+ primaryClass = {eess.AS},
34
+ url = {https://arxiv.org/abs/2512.00482}
35
+ }
36
+ ```
37
 
38
+ ## License
39
 
40
+ MIT.