Update app.py

app.py

@@ -3,96 +3,106 @@ import gradio as gr
 import torch
 import os
 import importlib.util
-
 from huggingface_hub import login, hf_hub_download
 
 RUN_DEMO = True
 
 if RUN_DEMO:
-
     from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor
     
     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-    
     print(f"Is CUDA available: {torch.cuda.is_available()}")
-    # True
-    #print(f"CUDA device: {torch.cuda.get_device_name(torch.cuda.current_device())}")
-    # Tesla T4
     
+    # Check token early
     auth_token = os.environ.get("hf_token")
-    print(auth_token)
     if not auth_token:
-        raise ValueError("Hugging Face token is missing! Add it as a secret.")
-
-    login(token=auth_token)
-
-    stt_module_path = hf_hub_download(repo_id="fahadqazi/private-code", filename="sindhi_stt_module.py", token=auth_token)
-
-    # Dynamically import the module
-    spec = importlib.util.spec_from_file_location("sindhi_stt_module", stt_module_path)
-    stt_module = importlib.util.module_from_spec(spec)
-    spec.loader.exec_module(stt_module)
-    
-    TranscriberClass = stt_module.Transcriber
-
-    Transcriber = TranscriberClass(
-        auth_token=auth_token
-    )
+        print("ERROR: Hugging Face token is missing!")
+        RUN_DEMO = False
+    else:
+        try:
+            login(token=auth_token)
+            
+            # Download module
+            stt_module_path = hf_hub_download(
+                repo_id="fahadqazi/private-code", 
+                filename="sindhi_stt_module.py", 
+                token=auth_token
+            )
+            
+            # Import module
+            spec = importlib.util.spec_from_file_location("sindhi_stt_module", stt_module_path)
+            stt_module = importlib.util.module_from_spec(spec)
+            spec.loader.exec_module(stt_module)
+            TranscriberClass = stt_module.Transcriber
+            
+            # LAZY LOADING: Don't initialize here, do it in function
+            print("Module loaded successfully, initializing on first use...")
+            
+        except Exception as e:
+            print(f"ERROR during initialization: {str(e)}")
+            import traceback
+            traceback.print_exc()
+            RUN_DEMO = False
 
+if RUN_DEMO:
+    # Initialize transcriber lazily
+    transcriber_instance = None
+    
+    def get_transcriber():
+        global transcriber_instance
+        if transcriber_instance is None:
+            print("Initializing Transcriber (first use)...")
+            transcriber_instance = TranscriberClass(auth_token=auth_token)
+            print("Transcriber initialized!")
+        return transcriber_instance
 
     if torch.cuda.is_available():
-        # Function to transcribe using the selected model
         @spaces.GPU(duration=60)
         def decorated_transcribe(audio_arrays):
-            return Transcriber.transcribe(audio_arrays)
+            return get_transcriber().transcribe(audio_arrays)
     else:
         def decorated_transcribe(audio_arrays):
-            return Transcriber.transcribe(audio_arrays)
+            return get_transcriber().transcribe(audio_arrays)
 
-    # def transcribe_wrapper(uploaded_file, youtube_link, remove_music):
     def transcribe_wrapper(uploaded_file, microphone, remove_music):
-        youtube_link = None
-        if youtube_link and youtube_link.strip() != "":
-            try:
-                clean_url = Transcriber.clean_youtube_url(youtube_link.strip())
-                audio_path = Transcriber.download_youtube_audio(clean_url)
-                is_youtube = True
-            except Exception as e:
-                return f"(YouTube download failed: {str(e)})"
-        elif microphone:
-            audio_path = microphone
-            is_youtube = False
-        elif uploaded_file:
-            audio_path = uploaded_file.name if hasattr(uploaded_file, "name") else uploaded_file
-            is_youtube = False
-        else:
-            return "Please upload a file, record from microphone, or enter a YouTube URL."
-        
-        audio_arrays = Transcriber.prepare_inputs(audio_path, is_youtube=is_youtube, use_demucs=remove_music)
-        print(audio_arrays)
-        return decorated_transcribe(audio_arrays)
+        try:
+            transcriber = get_transcriber()
+            
+            # ... rest of your code ...
+            youtube_link = None
+            if youtube_link and youtube_link.strip() != "":
+                # ... YouTube handling ...
+                pass
+            elif microphone:
+                audio_path = microphone
+                is_youtube = False
+            elif uploaded_file:
+                audio_path = uploaded_file.name if hasattr(uploaded_file, "name") else uploaded_file
+                is_youtube = False
+            else:
+                return "Please upload a file, record from microphone, or enter a YouTube URL."
+            
+            audio_arrays = transcriber.prepare_inputs(audio_path, is_youtube=is_youtube, use_demucs=remove_music)
+            return decorated_transcribe(audio_arrays)
+        except Exception as e:
+            return f"Error: {str(e)}"
     
     # Gradio Interface
     iface = gr.Interface(
         fn=transcribe_wrapper,
         inputs=[
             gr.File(file_types=[".wav", ".mp3", ".ogg", ".flac", ".mp4", ".mov", ".mkv"], label="Upload Audio or Video"),
-            # gr.Textbox(label="YouTube URL (optional)", placeholder="https://www.youtube.com/watch?v=..."),
             gr.Audio(sources="microphone", type="filepath", label="Or Record from Microphone"),
             gr.Checkbox(label="Remove background music / noise (slower, more accurate)", value=False)
         ],
-        # examples=[
-        #     [None, "https://www.youtube.com/watch?v=Dg2-4UX9NZU", False],
-        #     [None, "https://www.youtube.com/watch?v=MaE2zH9ZhQk", False]
-        # ],
         outputs="text",
         title="Sindhi Speech to Text",
-        # description="Upload an audio or video file (or paste a YouTube link) up to a few minutes long. Only one input is needed.",
         description="Upload an audio or video file up to a few minutes long. Only one input is needed.",
         cache_examples=False
     )
     
-    iface.launch(ssr=False)
+    print("Launching Gradio interface...")
+    iface.launch()
 
 else: