Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
## Usage
|
2 |
+
The model can be used directly (without a language model) as follows:
|
3 |
+
```python
|
4 |
+
import soundfile as sf
|
5 |
+
import torch
|
6 |
+
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
|
7 |
+
import argparse
|
8 |
+
def parse_transcription(wav_file):
|
9 |
+
# load pretrained model
|
10 |
+
processor = Wav2Vec2Processor.from_pretrained("addy88/wav2vec2-bhojpuri-stt")
|
11 |
+
model = Wav2Vec2ForCTC.from_pretrained("addy88/wav2vec2-bhojpuri-stt")
|
12 |
+
# load audio
|
13 |
+
audio_input, sample_rate = sf.read(wav_file)
|
14 |
+
# pad input values and return pt tensor
|
15 |
+
input_values = processor(audio_input, sampling_rate=sample_rate, return_tensors="pt").input_values
|
16 |
+
# INFERENCE
|
17 |
+
# retrieve logits & take argmax
|
18 |
+
logits = model(input_values).logits
|
19 |
+
predicted_ids = torch.argmax(logits, dim=-1)
|
20 |
+
# transcribe
|
21 |
+
transcription = processor.decode(predicted_ids[0], skip_special_tokens=True)
|
22 |
+
print(transcription)
|
23 |
+
```
|