Update app.py

app.py

@@ -9,7 +9,8 @@ import logging
 import os
 import spaces
 
-logging.basicConfig(level=logging.INFO)
+# Set up logging
+logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
 logger = logging.getLogger(__name__)
 
 def get_device():
@@ -18,7 +19,7 @@ def get_device():
     return torch.device("cpu")
 
 device = get_device()
-print(f"Using device: {device}")
+logger.info(f"Using device: {device}")
 
 model = None
 tokenizer = None
@@ -27,107 +28,132 @@ tokenizer = None
 def load_model():
     global model, tokenizer
     
-    print("Loading Orpheus model...")
+    logger.info("Loading Orpheus model...")
     model_name = "canopylabs/orpheus-3b-0.1-ft"
 
     hf_token = os.environ.get("HUGGINGFACE_TOKEN")
     if not hf_token:
         raise ValueError("HUGGINGFACE_TOKEN environment variable is not set")
 
-    login(token=hf_token)
-
-    snapshot_download(
-        repo_id=model_name,
-        use_auth_token=hf_token,
-        allow_patterns=[
-            "config.json",
-            "*.safetensors",
-            "model.safetensors.index.json",
-        ],
-        ignore_patterns=[
-            "optimizer.pt",
-            "pytorch_model.bin",
-            "training_args.bin",
-            "scheduler.pt",
-            "tokenizer.json",
-            "tokenizer_config.json",
-            "special_tokens_map.json",
-            "vocab.json",
-            "merges.txt",
-            "tokenizer.*"
-        ]
-    )
-
-    model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float32 if device.type == 'cpu' else torch.bfloat16)
-    model.to(device)
-    tokenizer = AutoTokenizer.from_pretrained(model_name)
-    print(f"Orpheus model and tokenizer loaded to {device}")
-
-def generate_podcast_script(api_key, content, duration, num_hosts):
-    genai.configure(api_key=api_key)
-    model = genai.GenerativeModel('gemini-2.5-pro-preview-03-25')
-    
-    prompt = f"""
-    Create a podcast script for {'one person' if num_hosts == 1 else 'two people'} discussing:
-    {content}
-    
-    Duration: {duration}. Include natural speech, humor, and occasional off-topic thoughts.
-    Use speech fillers like um, ah. Vary emotional tone.
-    
-    Format: {'Monologue' if num_hosts == 1 else 'Alternating dialogue'} without speaker labels.
-    Separate {'paragraphs' if num_hosts == 1 else 'lines'} with blank lines.
-    
-    Use emotion tags in angle brackets: <laugh>, <sigh>, <chuckle>, <cough>, <sniffle>, <groan>, <yawn>, <gasp>.
-    
-    Example: "I can't believe I stayed up all night <yawn> only to find out the meeting was canceled <groan>."
-    
-    Ensure content flows naturally and stays on topic. Match the script length to {duration}.
-    """
-    
-    response = model.generate_content(prompt)
-    return re.sub(r'[^a-zA-Z0-9\s.,?!<>]', '', response.text)
+    try:
+        login(token=hf_token)
+
+        snapshot_download(
+            repo_id=model_name,
+            use_auth_token=hf_token,
+            allow_patterns=[
+                "config.json",
+                "*.safetensors",
+                "model.safetensors.index.json",
+            ],
+            ignore_patterns=[
+                "optimizer.pt",
+                "pytorch_model.bin",
+                "training_args.bin",
+                "scheduler.pt",
+                "tokenizer.json",
+                "tokenizer_config.json",
+                "special_tokens_map.json",
+                "vocab.json",
+                "merges.txt",
+                "tokenizer.*"
+            ]
+        )
+
+        model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float32 if device.type == 'cpu' else torch.bfloat16)
+        model.to(device)
+        tokenizer = AutoTokenizer.from_pretrained(model_name)
+        logger.info(f"Orpheus model and tokenizer loaded to {device}")
+    except Exception as e:
+        logger.error(f"Error loading model: {str(e)}")
+        raise
+
+def generate_podcast_script(api_key, content, uploaded_file, duration, num_hosts):
+    try:
+        genai.configure(api_key=api_key)
+        model = genai.GenerativeModel('gemini-2.5-pro-preview-03-25')
+        
+        combined_content = content or ""
+        if uploaded_file:
+            file_content = uploaded_file.read().decode('utf-8')
+            combined_content += "\n" + file_content if combined_content else file_content
+        
+        prompt = f"""
+        Create a podcast script for {'one person' if num_hosts == 1 else 'two people'} discussing:
+        {combined_content}
+        
+        Duration: {duration}. Include natural speech, humor, and occasional off-topic thoughts.
+        Use speech fillers like um, ah. Vary emotional tone.
+        
+        Format: {'Monologue' if num_hosts == 1 else 'Alternating dialogue'} without speaker labels.
+        Separate {'paragraphs' if num_hosts == 1 else 'lines'} with blank lines.
+        
+        Use emotion tags in angle brackets: <laugh>, <sigh>, <chuckle>, <cough>, <sniffle>, <groan>, <yawn>, <gasp>.
+        
+        Example: "I can't believe I stayed up all night <yawn> only to find out the meeting was canceled <groan>."
+        
+        Ensure content flows naturally and stays on topic. Match the script length to {duration}.
+        """
+        
+        response = model.generate_content(prompt)
+        return re.sub(r'[^a-zA-Z0-9\s.,?!<>]', '', response.text)
+    except Exception as e:
+        logger.error(f"Error generating podcast script: {str(e)}")
+        raise
 
 @spaces.GPU()
 def text_to_speech(text, voice):
     global model, tokenizer
-    if model is None or tokenizer is None:
-        load_model()
-    
-    inputs = tokenizer(text, return_tensors="pt").to(device)
-    with torch.no_grad():
-        output = model.generate(**inputs, max_new_tokens=256)
-    mel = output[0].cpu().numpy()
-    audio = mel_to_audio(mel)
-    return audio
+    try:
+        if model is None or tokenizer is None:
+            load_model()
+        
+        inputs = tokenizer(text, return_tensors="pt").to(device)
+        with torch.no_grad():
+            output = model.generate(**inputs, max_new_tokens=256)
+        mel = output[0].cpu().numpy()
+        audio = mel_to_audio(mel)
+        return audio
+    except Exception as e:
+        logger.error(f"Error in text_to_speech: {str(e)}")
+        raise
 
 def mel_to_audio(mel):
     return np.zeros(24000, dtype=np.float32)  # Placeholder: 1 second of silence
 
 @spaces.GPU()
 def render_podcast(api_key, script, voice1, voice2, num_hosts):
-    lines = [line for line in script.split('\n') if line.strip()]
-    audio_segments = []
-
-    for i, line in enumerate(lines):
-        voice = voice1 if num_hosts == 1 or i % 2 == 0 else voice2
-        try:
-            audio = text_to_speech(line, voice)
-            audio_segments.append(audio)
-        except Exception as e:
-            logger.error(f"Error processing audio segment: {str(e)}")
-
-    if not audio_segments:
-        logger.warning("No valid audio segments were generated.")
-        return (24000, np.zeros(24000, dtype=np.float32))
-
-    podcast_audio = np.concatenate(audio_segments)
-    return (24000, podcast_audio)
+    try:
+        lines = [line for line in script.split('\n') if line.strip()]
+        audio_segments = []
+
+        for i, line in enumerate(lines):
+            voice = voice1 if num_hosts == 1 or i % 2 == 0 else voice2
+            try:
+                audio = text_to_speech(line, voice)
+                audio_segments.append(audio)
+            except Exception as e:
+                logger.error(f"Error processing audio segment: {str(e)}")
+
+        if not audio_segments:
+            logger.warning("No valid audio segments were generated.")
+            return (24000, np.zeros(24000, dtype=np.float32))
+
+        podcast_audio = np.concatenate(audio_segments)
+        return (24000, podcast_audio)
+    except Exception as e:
+        logger.error(f"Error rendering podcast: {str(e)}")
+        raise
 
 with gr.Blocks() as demo:
     gr.Markdown("# AI Podcast Generator")
     
     api_key_input = gr.Textbox(label="Enter your Gemini API Key", type="password")
-    content_input = gr.Textbox(label="Paste your content")
+    
+    with gr.Row():
+        content_input = gr.Textbox(label="Paste your content (optional)")
+        document_upload = gr.File(label="Upload Document (optional)")
+    
     duration = gr.Radio(["1-5 min", "5-10 min", "10-15 min"], label="Estimated podcast duration")
     num_hosts = gr.Radio([1, 2], label="Number of podcast hosts", value=2)
     
@@ -142,7 +168,7 @@ with gr.Blocks() as demo:
     audio_output = gr.Audio(label="Generated Podcast")
     
     generate_btn.click(generate_podcast_script, 
-                       inputs=[api_key_input, content_input, duration, num_hosts], 
+                       inputs=[api_key_input, content_input, document_upload, duration, num_hosts], 
                        outputs=script_output)
     
     render_btn.click(render_podcast, 
@@ -154,5 +180,8 @@ with gr.Blocks() as demo:
                      outputs=[voice2_select])
 
 if __name__ == "__main__":
-    load_model()
-    demo.launch()
+    try:
+        load_model()
+        demo.launch()
+    except Exception as e:
+        logger.error(f"Error launching the application: {str(e)}")