Update README.md
Browse files
README.md
CHANGED
|
@@ -9,3 +9,38 @@ tags:
|
|
| 9 |
- viet-asr
|
| 10 |
---
|
| 11 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
- viet-asr
|
| 10 |
---
|
| 11 |
|
| 12 |
+
## Usage
|
| 13 |
+
Use below Python to do ASR task:
|
| 14 |
+
```python
|
| 15 |
+
import torch
|
| 16 |
+
import librosa
|
| 17 |
+
from transformers import WhisperProcessor, WhisperForConditionalGeneration
|
| 18 |
+
|
| 19 |
+
# Load model and processor
|
| 20 |
+
processor = WhisperProcessor.from_pretrained("danhtran2mind/Vi-Whisper-tiny-finetuning")
|
| 21 |
+
model = WhisperForConditionalGeneration.from_pretrained("danhtran2mind/Vi-Whisper-tiny-finetuning")
|
| 22 |
+
model.config.forced_decoder_ids = None
|
| 23 |
+
# Move model to GPU if available
|
| 24 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 25 |
+
model.to(device)
|
| 26 |
+
|
| 27 |
+
# Load audio file (replace 'audio.wav' with your audio file path)
|
| 28 |
+
audio_path = "<audio_path>"
|
| 29 |
+
audio, sr = librosa.load(audio_path, sr=16000)
|
| 30 |
+
|
| 31 |
+
# Preprocess audio
|
| 32 |
+
inputs = processor(audio, sampling_rate=16000, return_tensors="pt").to(device)
|
| 33 |
+
|
| 34 |
+
# Perform inference with max_length and language
|
| 35 |
+
with torch.no_grad():
|
| 36 |
+
generated_ids = model.generate(
|
| 37 |
+
inputs["input_features"],
|
| 38 |
+
max_length=448,
|
| 39 |
+
)
|
| 40 |
+
|
| 41 |
+
# Decode the output
|
| 42 |
+
transcription = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
| 43 |
+
|
| 44 |
+
# Print the transcription
|
| 45 |
+
print("Transcription:\n", transcription)
|
| 46 |
+
```
|