Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,33 @@
|
|
| 1 |
---
|
| 2 |
license: mit
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: mit
|
| 3 |
+
datasets:
|
| 4 |
+
- mozilla-foundation/common_voice_13_0
|
| 5 |
+
language:
|
| 6 |
+
- ca
|
| 7 |
+
- cs
|
| 8 |
+
- gl
|
| 9 |
+
- hu
|
| 10 |
+
- pl
|
| 11 |
+
- ta
|
| 12 |
+
- th
|
| 13 |
+
- uk
|
| 14 |
---
|
| 15 |
+
|
| 16 |
+
## About
|
| 17 |
+
|
| 18 |
+
Multilingual Distilwhisper allows for better ASR performance in target languages by adding lightweight CLSR modules on top of whisper-small.
|
| 19 |
+
These modules are trained on a mix of cross-entropy (ASR) and knowledge distillation losses, where whisper-large-v2 is used as teacher.
|
| 20 |
+
|
| 21 |
+
## Inference
|
| 22 |
+
|
| 23 |
+
Loader will be made available soon at https://github.com/naver
|
| 24 |
+
|
| 25 |
+
## Citation (submitted to ICASSP 2024)
|
| 26 |
+
```
|
| 27 |
+
@article{ferraz2023distilwhisper,
|
| 28 |
+
title={DistilWhisper: Efficient Distillation of Multi-task Speech Models via Language-Specific Experts},
|
| 29 |
+
author={Ferraz, Thomas Palmeira and Boito, Marcely Zanon and Brun, Caroline and Nikoulina, Vassilina},
|
| 30 |
+
journal={arXiv preprint arXiv:2311.01070},
|
| 31 |
+
year={2023}
|
| 32 |
+
}
|
| 33 |
+
```
|