README.md
Browse files
README.md
ADDED
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
datasets:
|
4 |
+
- ASCEND
|
5 |
+
language:
|
6 |
+
- zh
|
7 |
+
metrics:
|
8 |
+
- cer
|
9 |
+
tags:
|
10 |
+
- audio
|
11 |
+
- automatic-speech-recognition
|
12 |
+
- speech
|
13 |
+
- xlsr-fine-tuning-week
|
14 |
+
---
|
15 |
+
|
16 |
+
|
17 |
+
## inference
|
18 |
+
|
19 |
+
The model can be used directly (without a language model) as follows...
|
20 |
+
|
21 |
+
Using the [HuggingSound](https://github.com/jonatasgrosman/huggingsound) library:
|
22 |
+
|
23 |
+
```python
|
24 |
+
from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
|
25 |
+
from datasets import load_dataset
|
26 |
+
import torch
|
27 |
+
import torchaudio
|
28 |
+
|
29 |
+
|
30 |
+
|
31 |
+
# load model and processor
|
32 |
+
processor = Wav2Vec2Processor.from_pretrained("gymeee/demo_code_switching")
|
33 |
+
model = Wav2Vec2ForCTC.from_pretrained("gymeee/demo_code_switching")
|
34 |
+
|
35 |
+
# load speech
|
36 |
+
speech_array, sampling_rate = torchaudio.load("speech.wav")
|
37 |
+
# tokenize
|
38 |
+
input_values = processor(speech_array[0], return_tensors="pt", padding="longest").input_values # Batch size 1
|
39 |
+
|
40 |
+
# retrieve logits
|
41 |
+
logits = model(input_values).logits
|
42 |
+
|
43 |
+
# take argmax and decode
|
44 |
+
predicted_ids = torch.argmax(logits, dim=-1)
|
45 |
+
transcription = processor.batch_decode(predicted_ids)
|
46 |
+
|
47 |
+
print(transcription)
|