ivangtorre
commited on
Commit
•
7044354
1
Parent(s):
71acad7
Update README.md
Browse files
README.md
CHANGED
@@ -33,6 +33,38 @@ transcription = processor.batch_decode(predicted_ids)
|
|
33 |
print("HF prediction: ", transcription)
|
34 |
```
|
35 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
36 |
## Citation
|
37 |
|
38 |
```bibtex
|
|
|
33 |
print("HF prediction: ", transcription)
|
34 |
```
|
35 |
|
36 |
+
|
37 |
+
This code snipnet shows how to Evaluate the wav2vec2-xlsr-300m-quechua in [Second Americas NLP 2022 Quechua dev set](https://huggingface.co/datasets/ivangtorre/second_americas_nlp_2022)
|
38 |
+
|
39 |
+
```python
|
40 |
+
from datasets import load_dataset
|
41 |
+
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
|
42 |
+
import torch
|
43 |
+
from jiwer import cer
|
44 |
+
import torch.nn.functional as F
|
45 |
+
|
46 |
+
|
47 |
+
#librispeech_eval = load_dataset("ivangtorre/second_americas_nlp_2022", split="validation")
|
48 |
+
librispeech_eval = load_dataset("ivangtorre/second_americas_nlp_2022", split="validation")
|
49 |
+
|
50 |
+
model = Wav2Vec2ForCTC.from_pretrained("ivangtorre/wav2vec2-xlsr-300m-quechua")
|
51 |
+
processor = Wav2Vec2Processor.from_pretrained("ivangtorre/wav2vec2-xlsr-300m-quechua")
|
52 |
+
|
53 |
+
def map_to_pred(batch):
|
54 |
+
wav = batch["audio"][0]["array"]
|
55 |
+
feats = torch.from_numpy(wav).float()
|
56 |
+
feats = F.layer_norm(feats, feats.shape) # Normalization performed during finetuning
|
57 |
+
feats = torch.unsqueeze(feats, 0)
|
58 |
+
logits = model(feats).logits
|
59 |
+
predicted_ids = torch.argmax(logits, dim=-1)
|
60 |
+
batch["transcription"] = processor.batch_decode(predicted_ids)
|
61 |
+
return batch
|
62 |
+
|
63 |
+
result = librispeech_eval.map(map_to_pred, batched=True, batch_size=1)
|
64 |
+
|
65 |
+
print("CER:", cer(result["source_processed"], result["transcription"]))
|
66 |
+
```
|
67 |
+
|
68 |
## Citation
|
69 |
|
70 |
```bibtex
|