Spaces:
Paused
Paused
Muhammed_Kotb1
commited on
Commit
•
5dd8287
1
Parent(s):
214097c
test ziad model
Browse files
app.py
CHANGED
@@ -1,7 +1,55 @@
|
|
1 |
-
import
|
|
|
|
|
2 |
|
3 |
-
|
4 |
-
|
|
|
|
|
|
|
5 |
|
6 |
-
|
7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import torchaudio
|
3 |
+
from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC, Wav2Vec2CTCTokenizer
|
4 |
|
5 |
+
# Load the Arabic-specific processor and model
|
6 |
+
model_name = "Zaid/wav2vec2-large-xlsr-53-arabic-egyptian"
|
7 |
+
tokenizer = Wav2Vec2CTCTokenizer.from_pretrained(model_name)
|
8 |
+
processor = Wav2Vec2Processor.from_pretrained(model_name, tokenizer=tokenizer)
|
9 |
+
model = Wav2Vec2ForCTC.from_pretrained(model_name)
|
10 |
|
11 |
+
def transcribe(audio_file):
|
12 |
+
try:
|
13 |
+
# Load the audio file
|
14 |
+
print("Loading audio file...")
|
15 |
+
audio_input, sr = torchaudio.load(audio_file)
|
16 |
+
print(f"Audio loaded: {audio_input.shape}, Sample rate: {sr}")
|
17 |
+
|
18 |
+
# Resample if needed
|
19 |
+
if sr != 16000:
|
20 |
+
print("Resampling audio...")
|
21 |
+
resampler = torchaudio.transforms.Resample(orig_freq=sr, new_freq=16000)
|
22 |
+
audio_input = resampler(audio_input)
|
23 |
+
sr = 16000
|
24 |
+
|
25 |
+
print(f"Audio shape after resampling: {audio_input.shape}, Sample rate: {sr}")
|
26 |
+
|
27 |
+
# Convert tensor to numpy array
|
28 |
+
audio_input = audio_input[0].numpy()
|
29 |
+
|
30 |
+
# Process audio input
|
31 |
+
print("Processing audio input...")
|
32 |
+
input_values = processor(audio_input, return_tensors="pt", sampling_rate=sr).input_values
|
33 |
+
|
34 |
+
# Run model inference
|
35 |
+
print("Running model inference...")
|
36 |
+
with torch.no_grad():
|
37 |
+
logits = model(input_values).logits
|
38 |
+
|
39 |
+
# Decode transcription
|
40 |
+
print("Decoding transcription...")
|
41 |
+
predicted_ids = torch.argmax(logits, dim=-1)
|
42 |
+
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True, clean_up_tokenization_spaces=True)
|
43 |
+
return transcription[0]
|
44 |
+
except Exception as e:
|
45 |
+
print(f"An error occurred: {e}")
|
46 |
+
return None
|
47 |
+
|
48 |
+
# Transcribe the audio file
|
49 |
+
transcription = transcribe("sidiali.wav")
|
50 |
+
if transcription:
|
51 |
+
print(transcription.encode('utf-8').decode('utf-8'))
|
52 |
+
# Save the transcription to a file
|
53 |
+
with open("transcription.txt", "w", encoding="utf-8") as f:
|
54 |
+
f.write(transcription)
|
55 |
+
print("Transcription saved to transcription.txt")
|