ivangtorre
commited on
Commit
•
dd9561c
1
Parent(s):
5787097
Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,28 @@
|
|
1 |
-
---
|
2 |
-
license: cc-by-4.0
|
3 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: cc-by-4.0
|
3 |
+
---
|
4 |
+
|
5 |
+
## Usage
|
6 |
+
|
7 |
+
The model can be used directly (without a language model) as follows:
|
8 |
+
|
9 |
+
```python
|
10 |
+
from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
|
11 |
+
import torch
|
12 |
+
import torchaudio
|
13 |
+
|
14 |
+
# load model and processor
|
15 |
+
processor = Wav2Vec2Processor.from_pretrained("ivangtorre/wav2vec2-xls-r-300m-quechua")
|
16 |
+
model = Wav2Vec2ForCTC.from_pretrained("ivangtorre/wav2vec2-xls-r-300m-quechua")
|
17 |
+
|
18 |
+
# load dummy dataset and read soundfiles
|
19 |
+
file = torchaudio.load("quechua000573.wav")
|
20 |
+
|
21 |
+
# retrieve logits
|
22 |
+
logits = model(file[0]).logits
|
23 |
+
|
24 |
+
# take argmax and decode
|
25 |
+
predicted_ids = torch.argmax(logits, dim=-1)
|
26 |
+
transcription = processor.batch_decode(predicted_ids)
|
27 |
+
print("HF prediction: ", transcription)
|
28 |
+
```
|