Iliass Lasri commited on
Commit ·
a9e7ffa
1
Parent(s): eff87e0
add layer
Browse files
README.md
CHANGED
|
@@ -76,17 +76,17 @@ We trained quantizers across different encoders, codebook sizes, and augmentatio
|
|
| 76 |
- **All augmentations, single** — all augmentations are enabled, but only one randomly chosen augmentation is applied per sample.
|
| 77 |
- **No extra augmentations, single** — only the baseline augmentations (from the original paper) are used, with one applied per sample.
|
| 78 |
|
| 79 |
-
| Encoder | Codebook | Augmentation Strategy |
|
| 80 |
-
|:---|:---:|:---|
|
| 81 |
-
| HuBERT | 500 | All augmentations, chained |
|
| 82 |
-
|
|
| 83 |
-
|
|
| 84 |
-
| | | |
|
| 85 |
-
| SpidR | 256 | No extra augmentations, single |
|
| 86 |
-
|
|
| 87 |
-
| | | |
|
| 88 |
-
| DinoSR (original) | 256 | All augmentations, chained |
|
| 89 |
-
| DinoSR (reproduced) | 256 | All augmentations, chained |
|
| 90 |
|
| 91 |
## Links
|
| 92 |
- Paper: [Algayres et al., Interspeech 2023](https://aclanthology.org/2023.iwslt-1.46/)
|
|
|
|
| 76 |
- **All augmentations, single** — all augmentations are enabled, but only one randomly chosen augmentation is applied per sample.
|
| 77 |
- **No extra augmentations, single** — only the baseline augmentations (from the original paper) are used, with one applied per sample.
|
| 78 |
|
| 79 |
+
| Encoder | Layer | Codebook | Augmentation Strategy |
|
| 80 |
+
|:---|:---:|:---:|:---|
|
| 81 |
+
| HuBERT | 6 | 500 | All augmentations, chained |
|
| 82 |
+
| | | | All augmentations, single |
|
| 83 |
+
| | | | No extra augmentations, single |
|
| 84 |
+
| | | | |
|
| 85 |
+
| SpidR | 6 | 256 | No extra augmentations, single |
|
| 86 |
+
| | | | All augmentations, chained |
|
| 87 |
+
| | | | |
|
| 88 |
+
| DinoSR (original) | 5 | 256 | All augmentations, chained |
|
| 89 |
+
| DinoSR (reproduced) | 5 | 256 | All augmentations, chained |
|
| 90 |
|
| 91 |
## Links
|
| 92 |
- Paper: [Algayres et al., Interspeech 2023](https://aclanthology.org/2023.iwslt-1.46/)
|