AIvry commited on
Commit
a41124a
·
verified ·
1 Parent(s): 9ae59a2

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +10 -11
app.py CHANGED
@@ -215,9 +215,9 @@ def create_interface():
215
  - `manifest_canonical.json`: File mapping and processing details
216
 
217
  ### Score Interpretation
218
- - **NaN values**: Appear in frames where fewer than 2 speakers are active
219
  - **Valid scores**: Only computed when at least 2 speakers are active in a frame
220
- - **Time resolution**: 20ms frames (configurable in code)
 
221
 
222
  ## Available Models
223
 
@@ -226,10 +226,10 @@ def create_interface():
226
  | `raw` | Raw waveform features | N/A | Baseline comparison |
227
  | `wavlm` | WavLM Large | 24 | Strong performance |
228
  | `wav2vec2` | Wav2Vec2 Large | 24 | Best overall performance |
229
- | `hubert` | HuBERT Large | 24 | Good for speech |
230
- | `wavlm_base` | WavLM Base | 12 | Faster processing |
231
  | `wav2vec2_base` | Wav2Vec2 Base | 12 | Faster, good quality |
232
- | `hubert_base` | HuBERT Base | 12 | Faster processing |
233
  | `wav2vec2_xlsr` | Wav2Vec2 XLSR-53 | 24 | Multilingual |
234
 
235
  ## Parameters
@@ -253,12 +253,11 @@ def create_interface():
253
  If you use MAPSS, please cite:
254
 
255
  ```bibtex
256
- @article{Ivry2025MAPSS,
257
- title = {MAPSS: Manifold-based Assessment of Perceptual Source Separation},
258
- author = {Ivry, Amir and Cornell, Samuele and Watanabe, Shinji},
259
- journal = {arXiv preprint arXiv:2509.09212},
260
- year = {2025},
261
- url = {https://arxiv.org/abs/2509.09212}
262
  }
263
  ```
264
 
 
215
  - `manifest_canonical.json`: File mapping and processing details
216
 
217
  ### Score Interpretation
 
218
  - **Valid scores**: Only computed when at least 2 speakers are active in a frame
219
+ - **NaN values**: Appear for non-active speakers, or when fewer than 2 speakers are active in the frame.
220
+ - **Time resolution**: 20ms frames
221
 
222
  ## Available Models
223
 
 
226
  | `raw` | Raw waveform features | N/A | Baseline comparison |
227
  | `wavlm` | WavLM Large | 24 | Strong performance |
228
  | `wav2vec2` | Wav2Vec2 Large | 24 | Best overall performance |
229
+ | `hubert` | HuBERT Large | 24 | |
230
+ | `wavlm_base` | WavLM Base | 12 | |
231
  | `wav2vec2_base` | Wav2Vec2 Base | 12 | Faster, good quality |
232
+ | `hubert_base` | HuBERT Base | 12 | |
233
  | `wav2vec2_xlsr` | Wav2Vec2 XLSR-53 | 24 | Multilingual |
234
 
235
  ## Parameters
 
253
  If you use MAPSS, please cite:
254
 
255
  ```bibtex
256
+ @article{ivry2025mapss,
257
+ title={MAPSS: Manifold-based Assessment of Perceptual Source Separation},
258
+ author={Ivry, Amir and Cornell, Samuele and Watanabe, Shinji},
259
+ journal={arXiv preprint arXiv:2509.09212},
260
+ year={2025}
 
261
  }
262
  ```
263