Update app.py

app.py

@@ -5,8 +5,12 @@ import shutil
 import subprocess
 from faster_whisper import WhisperModel
 
+# 🔤 Hindi Script Fix
+from indic_transliteration import sanscript
+from indic_transliteration.sanscript import transliterate
+
 # ===============================
-# 1. Whisper Model
+# 1. Whisper Model (Lazy Load)
 # ===============================
 model = None
 
@@ -21,12 +25,11 @@ def load_model():
 # ===============================
 # 2. FFmpeg Path
 # ===============================
-def get_ffmpeg_path():
-    path = shutil.which("ffmpeg")
-    return path if path else "/usr/bin/ffmpeg"
+def get_ffmpeg():
+    return shutil.which("ffmpeg") or "/usr/bin/ffmpeg"
 
 # ===============================
-# 3. Extract Audio
+# 3. Video → Audio
 # ===============================
 def extract_audio(video_path):
     audio_path = "uploaded_audio.wav"
@@ -34,7 +37,7 @@ def extract_audio(video_path):
         os.remove(audio_path)
 
     cmd = [
-        get_ffmpeg_path(),
+        get_ffmpeg(),
         "-i", video_path,
         "-vn",
         "-ac", "1",
@@ -46,15 +49,14 @@ def extract_audio(video_path):
     return audio_path
 
 # ===============================
-# 4. Download Audio
+# 4. Download Audio from URL
 # ===============================
 def download_audio_from_url(url):
-    output = "url_audio.%(ext)s"
+    output = "url_audio"
 
     ydl_opts = {
         "format": "bestaudio/best",
         "outtmpl": output,
-        "ffmpeg_location": os.path.dirname(get_ffmpeg_path()),
         "postprocessors": [{
             "key": "FFmpegExtractAudio",
             "preferredcodec": "wav",
@@ -69,7 +71,18 @@ def download_audio_from_url(url):
     return "url_audio.wav"
 
 # ===============================
-# 5. Transcribe Function
+# 5. Hindi Script Normalizer
+# ===============================
+def normalize_script(text, lang):
+    if lang == "hi":
+        try:
+            return transliterate(text, sanscript.ARABIC, sanscript.DEVANAGARI)
+        except:
+            return text
+    return text
+
+# ===============================
+# 6. Main Transcribe Logic
 # ===============================
 def transcribe_media(url_input, file_input, language_choice):
 
@@ -89,14 +102,14 @@ def transcribe_media(url_input, file_input, language_choice):
             audio_path = download_audio_from_url(url_input)
 
         else:
-            return "⚠️ Please provide URL or Upload file."
+            return "⚠️ Please paste a link or upload a file."
 
         if not os.path.exists(audio_path):
             return "❌ Audio processing failed."
 
         model = load_model()
 
-        # Language Handling
+        # Language handling
         language = None if language_choice == "Auto Detect" else language_choice
 
         segments, info = model.transcribe(
@@ -107,16 +120,16 @@ def transcribe_media(url_input, file_input, language_choice):
         )
 
         detected_lang = info.language
+        raw_text = " ".join(seg.text for seg in segments)
+        final_text = normalize_script(raw_text, detected_lang)
 
-        text = " ".join(seg.text for seg in segments)
-
-        return f"🌍 Detected Language: {detected_lang}\n\n{text.strip()}"
+        return f"🌍 Detected Language: {detected_lang}\n\n{final_text.strip()}"
 
     except Exception as e:
         return f"❌ Error: {str(e)}"
 
 # ===============================
-# 6. UI
+# 7. UI
 # ===============================
 css = """
 .container {max-width: 900px; margin: auto;}
@@ -130,7 +143,10 @@ css = """
 with gr.Blocks(theme=gr.themes.Soft(), css=css) as demo:
     with gr.Column(elem_classes="container"):
         gr.Markdown("## 🚀 Universal Transcript Tool")
-        gr.Markdown("Supports YouTube, TikTok, Instagram, Facebook, Twitter/X")
+        gr.Markdown(
+            "Supports **YouTube, TikTok, Instagram, Facebook, Twitter/X**\n\n"
+            "Hindi output is always **Devanagari** 🇮🇳"
+        )
 
         with gr.Tabs():
             with gr.TabItem("🔗 Paste Link"):
@@ -148,16 +164,16 @@ with gr.Blocks(theme=gr.themes.Soft(), css=css) as demo:
         language_selector = gr.Dropdown(
             choices=[
                 "Auto Detect",
-                "en",  # English
-                "hi",  # Hindi
+                "hi",  # Hindi (Devanagari)
                 "ur",  # Urdu
-                "ar",  # Arabic
-                "fr",  # French
-                "de",  # German
-                "es",  # Spanish
-                "ru",  # Russian
-                "ja",  # Japanese
-                "zh"   # Chinese
+                "en",  # English
+                "ar",
+                "fr",
+                "de",
+                "es",
+                "ru",
+                "ja",
+                "zh"
             ],
             value="Auto Detect",
             label="🌍 Select Transcript Language"