Spaces:
Build error
Build error
Create train.py
Browse files
train.py
ADDED
@@ -0,0 +1,54 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from torch.utils.data import DataLoader
|
3 |
+
from datasets import load_dataset
|
4 |
+
from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
|
5 |
+
from transformers import Trainer, TrainingArguments
|
6 |
+
|
7 |
+
# تحميل بيانات SADA
|
8 |
+
dataset = load_dataset("m6011/sada2022")
|
9 |
+
|
10 |
+
# تحميل نموذج Wav2Vec2 لتحويل الصوت إلى نص (يمكنك تغييره إذا كنت تود استخدام نموذج آخر)
|
11 |
+
processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-large-xlsr-53")
|
12 |
+
model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-large-xlsr-53")
|
13 |
+
|
14 |
+
# معالجة البيانات - تحويل النص إلى رموز صوتية مناسبة (حسب النموذج المختار)
|
15 |
+
def preprocess_data(batch):
|
16 |
+
audio = batch["audio"]
|
17 |
+
inputs = processor(audio["array"], sampling_rate=audio["sampling_rate"], return_tensors="pt", padding=True)
|
18 |
+
batch["input_values"] = inputs.input_values[0]
|
19 |
+
batch["attention_mask"] = inputs.attention_mask[0]
|
20 |
+
|
21 |
+
# تحويل النص إلى رموز
|
22 |
+
with processor.as_target_processor():
|
23 |
+
batch["labels"] = processor(batch["ProcessedText"]).input_ids
|
24 |
+
return batch
|
25 |
+
|
26 |
+
# تطبيق المعالجة المسبقة على البيانات
|
27 |
+
dataset = dataset.map(preprocess_data, remove_columns=["audio", "ProcessedText"])
|
28 |
+
|
29 |
+
# إعدادات التدريب
|
30 |
+
training_args = TrainingArguments(
|
31 |
+
output_dir="./wav2vec2-saudi-tts",
|
32 |
+
group_by_length=True,
|
33 |
+
per_device_train_batch_size=4,
|
34 |
+
evaluation_strategy="steps",
|
35 |
+
num_train_epochs=3,
|
36 |
+
save_steps=400,
|
37 |
+
eval_steps=400,
|
38 |
+
logging_steps=400,
|
39 |
+
learning_rate=3e-4,
|
40 |
+
warmup_steps=500,
|
41 |
+
save_total_limit=2,
|
42 |
+
)
|
43 |
+
|
44 |
+
# إعداد المدرب
|
45 |
+
trainer = Trainer(
|
46 |
+
model=model,
|
47 |
+
args=training_args,
|
48 |
+
train_dataset=dataset["train"],
|
49 |
+
eval_dataset=dataset["test"],
|
50 |
+
tokenizer=processor.feature_extractor,
|
51 |
+
)
|
52 |
+
|
53 |
+
# بدء التدريب
|
54 |
+
trainer.train()
|