Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -8,10 +8,12 @@ from moviepy import VideoFileClip, TextClip, CompositeVideoClip
|
|
| 8 |
from arabic_reshaper import reshape
|
| 9 |
|
| 10 |
# --- الإعدادات ---
|
|
|
|
| 11 |
model = WhisperModel("large-v3", device="cpu", compute_type="int8")
|
| 12 |
|
| 13 |
def process_arabic_text(text):
|
| 14 |
if not text: return ""
|
|
|
|
| 15 |
return reshape(text)
|
| 16 |
|
| 17 |
def clean_color(color_str):
|
|
@@ -22,17 +24,18 @@ def clean_color(color_str):
|
|
| 22 |
return f'rgb({r},{g},{b})'
|
| 23 |
return color_str
|
| 24 |
|
| 25 |
-
def
|
| 26 |
if not video_path: return None, "الرجاء رفع فيديو."
|
| 27 |
|
| 28 |
-
#
|
| 29 |
-
segments, _ = model.transcribe(video_path, language="ar")
|
| 30 |
|
| 31 |
-
|
| 32 |
-
for
|
| 33 |
-
|
|
|
|
| 34 |
|
| 35 |
-
return pd.DataFrame(
|
| 36 |
|
| 37 |
def step_2_render_video(video_path, df_edited, font_selection, text_color, font_size, progress=gr.Progress()):
|
| 38 |
if video_path is None or df_edited is None: return None, "بيانات ناقصة."
|
|
@@ -40,56 +43,55 @@ def step_2_render_video(video_path, df_edited, font_selection, text_color, font_
|
|
| 40 |
safe_color = clean_color(text_color)
|
| 41 |
actual_font = font_selection if os.path.exists(font_selection) else "DejaVu-Sans-Bold"
|
| 42 |
|
| 43 |
-
output_path = "
|
| 44 |
video = VideoFileClip(video_path)
|
| 45 |
w, h = video.size
|
| 46 |
clips = [video]
|
| 47 |
|
| 48 |
-
|
| 49 |
-
segments_list = df_edited.values.tolist()
|
| 50 |
|
| 51 |
-
|
| 52 |
-
|
|
|
|
| 53 |
t_start = float(row[1])
|
| 54 |
t_end = float(row[2])
|
| 55 |
|
| 56 |
-
if not
|
| 57 |
|
| 58 |
-
|
| 59 |
|
| 60 |
-
# إنشاء
|
| 61 |
txt = TextClip(
|
| 62 |
-
text=
|
| 63 |
font_size=int(font_size),
|
| 64 |
color=safe_color,
|
| 65 |
stroke_color='black',
|
| 66 |
-
stroke_width=
|
| 67 |
font=actual_font,
|
| 68 |
-
method='
|
| 69 |
-
|
| 70 |
-
text_align='center'
|
| 71 |
-
).with_start(t_start).with_duration(max(0.1, t_end - t_start)).with_position(('center', int(h * 0.75)))
|
| 72 |
|
| 73 |
clips.append(txt)
|
| 74 |
|
|
|
|
| 75 |
final = CompositeVideoClip(clips, size=(w, h))
|
| 76 |
final.write_videofile(output_path, codec="libx264", audio_codec="aac", fps=video.fps, logger='bar')
|
| 77 |
-
return output_path, "تم إنتاج الفيديو بن
|
| 78 |
|
| 79 |
# --- الواجهة ---
|
| 80 |
with gr.Blocks() as app:
|
| 81 |
-
gr.Markdown("## 🎬
|
| 82 |
with gr.Row():
|
| 83 |
v_in = gr.Video(); v_out = gr.Video()
|
| 84 |
with gr.Row():
|
| 85 |
font_opt = gr.Dropdown(choices=["arialbd.ttf"], value="arialbd.ttf", label="الخط")
|
| 86 |
-
color_opt = gr.ColorPicker(value="#
|
| 87 |
-
size_opt = gr.Slider(
|
| 88 |
|
| 89 |
-
btn_1 = gr.Button("1.
|
| 90 |
-
btn_2 = gr.Button("2.
|
| 91 |
|
| 92 |
-
btn_1.click(
|
| 93 |
btn_2.click(step_2_render_video, [v_in, table, font_opt, color_opt, size_opt], [v_out, status])
|
| 94 |
|
| 95 |
app.launch()
|
|
|
|
| 8 |
from arabic_reshaper import reshape
|
| 9 |
|
| 10 |
# --- الإعدادات ---
|
| 11 |
+
# استخدمنا cpu و int8 لضمان العمل على أغلب الأجهزة، يمكن تغييره لـ cuda إذا توفر GPU
|
| 12 |
model = WhisperModel("large-v3", device="cpu", compute_type="int8")
|
| 13 |
|
| 14 |
def process_arabic_text(text):
|
| 15 |
if not text: return ""
|
| 16 |
+
# معالجة النصوص العربية لتعرض بشكل صحيح (من اليمين لليسار وبدون حروف مقطعة)
|
| 17 |
return reshape(text)
|
| 18 |
|
| 19 |
def clean_color(color_str):
|
|
|
|
| 24 |
return f'rgb({r},{g},{b})'
|
| 25 |
return color_str
|
| 26 |
|
| 27 |
+
def step_1_extract_words(video_path, progress=gr.Progress()):
|
| 28 |
if not video_path: return None, "الرجاء رفع فيديو."
|
| 29 |
|
| 30 |
+
# تفعيل word_timestamps=True هو السر لاستخراج توقيت كل كلمة
|
| 31 |
+
segments, _ = model.transcribe(video_path, word_timestamps=True, language="ar")
|
| 32 |
|
| 33 |
+
words_data = []
|
| 34 |
+
for segment in segments:
|
| 35 |
+
for word in segment.words:
|
| 36 |
+
words_data.append([word.word.strip(), round(word.start, 2), round(word.end, 2)])
|
| 37 |
|
| 38 |
+
return pd.DataFrame(words_data, columns=["الكلمة", "البداية", "النهاية"]), "تم استخراج الكلمات بدقة!"
|
| 39 |
|
| 40 |
def step_2_render_video(video_path, df_edited, font_selection, text_color, font_size, progress=gr.Progress()):
|
| 41 |
if video_path is None or df_edited is None: return None, "بيانات ناقصة."
|
|
|
|
| 43 |
safe_color = clean_color(text_color)
|
| 44 |
actual_font = font_selection if os.path.exists(font_selection) else "DejaVu-Sans-Bold"
|
| 45 |
|
| 46 |
+
output_path = "final_word_by_word.mp4"
|
| 47 |
video = VideoFileClip(video_path)
|
| 48 |
w, h = video.size
|
| 49 |
clips = [video]
|
| 50 |
|
| 51 |
+
words_list = df_edited.values.tolist()
|
|
|
|
| 52 |
|
| 53 |
+
# تحويل كل صف في الجدول (كلمة) إلى Clip مستقل يظهر في وقته
|
| 54 |
+
for row in words_list:
|
| 55 |
+
word_text = str(row[0])
|
| 56 |
t_start = float(row[1])
|
| 57 |
t_end = float(row[2])
|
| 58 |
|
| 59 |
+
if not word_text.strip(): continue
|
| 60 |
|
| 61 |
+
clean_word = process_arabic_text(word_text)
|
| 62 |
|
| 63 |
+
# إنشاء كليب للكلمة الواحدة
|
| 64 |
txt = TextClip(
|
| 65 |
+
text=clean_word,
|
| 66 |
font_size=int(font_size),
|
| 67 |
color=safe_color,
|
| 68 |
stroke_color='black',
|
| 69 |
+
stroke_width=2,
|
| 70 |
font=actual_font,
|
| 71 |
+
method='label', # 'label' أفضل للكلمات المنفردة للحفاظ على جودتها
|
| 72 |
+
).with_start(t_start).with_duration(max(0.1, t_end - t_start)).with_position(('center', int(h * 0.5))) # وضعها في منتصف الشاشة تقريباً
|
|
|
|
|
|
|
| 73 |
|
| 74 |
clips.append(txt)
|
| 75 |
|
| 76 |
+
# دمج كل الكلمات فوق الفيديو الأصلي
|
| 77 |
final = CompositeVideoClip(clips, size=(w, h))
|
| 78 |
final.write_videofile(output_path, codec="libx264", audio_codec="aac", fps=video.fps, logger='bar')
|
| 79 |
+
return output_path, "تم إنتاج الفيديو بنظام كلمة بكلمة!"
|
| 80 |
|
| 81 |
# --- الواجهة ---
|
| 82 |
with gr.Blocks() as app:
|
| 83 |
+
gr.Markdown("## 🎬 Word-by-Word Captioner (Like Pro Apps)")
|
| 84 |
with gr.Row():
|
| 85 |
v_in = gr.Video(); v_out = gr.Video()
|
| 86 |
with gr.Row():
|
| 87 |
font_opt = gr.Dropdown(choices=["arialbd.ttf"], value="arialbd.ttf", label="الخط")
|
| 88 |
+
color_opt = gr.ColorPicker(value="#00FF00", label="لون الكلمة (مثلاً أخضر)")
|
| 89 |
+
size_opt = gr.Slider(50, 250, value=120, label="حجم الكلمة")
|
| 90 |
|
| 91 |
+
btn_1 = gr.Button("1. تحليل الكلمات المنطوقة"); table = gr.Dataframe(interactive=True)
|
| 92 |
+
btn_2 = gr.Button("2. إنشاء الفيديو النهائي"); status = gr.Textbox()
|
| 93 |
|
| 94 |
+
btn_1.click(step_1_extract_words, [v_in], [table, status])
|
| 95 |
btn_2.click(step_2_render_video, [v_in, table, font_opt, color_opt, size_opt], [v_out, status])
|
| 96 |
|
| 97 |
app.launch()
|