Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,43 @@
|
|
1 |
---
|
2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
+
language:
|
4 |
+
- en
|
5 |
+
library_name: fairseq
|
6 |
+
pipeline_tag: automatic-speech-recognition
|
7 |
+
inference: false
|
8 |
---
|
9 |
+
|
10 |
+
|
11 |
+
<br>
|
12 |
+
<br>
|
13 |
+
|
14 |
+
# ARMHuBERT Model Card
|
15 |
+
|
16 |
+
This repo contains the models from our paper [**Recycle-and-Distill: Universal Compression Strategy for Transformer-based Speech SSL Models with Attention Map Reusing and Masking Distillation**](https://arxiv.org/abs/2305.11685), INTERSPEECH 2023.
|
17 |
+
|
18 |
+
|
19 |
+
## Model details
|
20 |
+
|
21 |
+
**Model type:**
|
22 |
+
ARMHuBERT is an open-source speech SSL model distilled from HuBERT-Base, by attention map reusing and masking distillation.
|
23 |
+
We also provide the model checkpoints of MaskHuBERT (without attention map reusing) and ARMwavLM (wavLM-Base teacher).
|
24 |
+
|
25 |
+
- Attention Map Reusing: Reuse previous layer's attention map to remove key & query parameters in Transformer.
|
26 |
+
- Masking Distillation: Masking distillation treating masked frames and unmasked frames separately.
|
27 |
+
|
28 |
+
**License:**
|
29 |
+
Apache 2.0 License
|
30 |
+
|
31 |
+
**Where to send questions or comments about the model:**
|
32 |
+
https://github.com/sungnyun/ARMHuBERT/issues
|
33 |
+
|
34 |
+
|
35 |
+
## Training dataset
|
36 |
+
Pretraining data: [LibriSpeech](https://www.openslr.org/12)
|
37 |
+
- ``[ModelName]-100h.ckpt``: train-clean-100
|
38 |
+
- ``[ModelName]-960h.ckpt``: train-clean-100 + train-clean-360 + train-other-500
|
39 |
+
|
40 |
+
|
41 |
+
<br>
|
42 |
+
|
43 |
+
More detials are in our github, https://github.com/sungnyun/ARMHuBERT.
|