Added revisions to the translation app

video_transcription/ai_transcriber_v2/transcriber.py

@@ -115,7 +115,28 @@ def save_as_srt(result, output_path):
             f.write(f"{text}\n\n")
     print(f"SRT saved to: {output_path}")
 
-def transcribe_audio(audio_path, model_size="auto", language=None):
+def load_whisper_model(model_size="auto"):
+    """
+    Loads and returns the Whisper model.
+    """
+    check_gpu_health()
+
+    if model_size == "auto":
+        model_size = get_optimal_model_size()
+        print(f"Auto-selected model: '{model_size}'")
+
+    print(f"Loading Whisper model ('{model_size}')...")
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    print(f"Using device: {device}")
+
+    try:
+        model = whisper.load_model(model_size, device=device)
+        return model
+    except Exception as e:
+        print(f"Error loading model: {e}")
+        sys.exit(1)
+
+def transcribe_audio(audio_path, model_size="auto", language=None, loaded_model=None):
     """
     Transcribes an audio file using OpenAI's Whisper model.
 
@@ -123,6 +144,7 @@ def transcribe_audio(audio_path, model_size="auto", language=None):
         audio_path (str): Path to the input audio file.
         model_size (str): Size of the Whisper model to use. If "auto", selects based on VRAM.
         language (str, optional): Language code (e.g., "en", "fr", "es"). If None, auto-detects.
+        loaded_model (object, optional): Pre-loaded Whisper model object.
 
     Returns:
         dict: The full transcription result containing segments and text.
@@ -130,37 +152,14 @@ def transcribe_audio(audio_path, model_size="auto", language=None):
     if not os.path.exists(audio_path):
         raise FileNotFoundError(f"Audio file not found: {audio_path}")
 
-    # Run health check once
-    check_gpu_health()
-
-    # Determine model size if auto
-    if model_size == "auto":
-        model_size = get_optimal_model_size()
-        print(f"Auto-selected model: '{model_size}'")
-
-    print(f"Loading Whisper model ('{model_size}')...")
-    
-    # Check for GPU availability
-    device = "cuda" if torch.cuda.is_available() else "cpu"
-    print(f"Using device: {device}")
-
-    try:
-        model = whisper.load_model(model_size, device=device)
-    except RuntimeError as e:
-        if "out of memory" in str(e).lower():
-            print("Error: GPU Out of Memory. Try using a smaller model size.")
-        else:
-            print(f"Error loading model: {e}")
-        sys.exit(1)
-    except Exception as e:
-        print(f"Error loading model: {e}")
-        sys.exit(1)
+    model = loaded_model
+    if model is None:
+        model = load_whisper_model(model_size)
 
     print(f"Transcribing {audio_path}...")
     try:
-        # fp16=False is needed for CPU, but we can let whisper handle defaults usually.
-        # language=None allows auto-detection.
-        result = model.transcribe(audio_path, language=language)
+        # Enable verbose=True to show progress in terminal
+        result = model.transcribe(audio_path, language=language, verbose=True)
         print("Transcription complete.")
         return result
     except Exception as e: