Update app.py

app.py

@@ -1,237 +1,211 @@
 #!/usr/bin/env python3
 """
-Complete Working MiniCPM-o-2.6 Voiceover Generator
-No Docker needed - Direct Python run
+MiniCPM-o-2.6 Text Generator with Gradio UI
+Convert technical prompts to natural voiceover text
+No TTS - Pure MiniCPM text generation with Web Interface
 """
 
 import gradio as gr
 import torch
+import csv
 import tempfile
-import os
-import subprocess
-import sys
 from pathlib import Path
 import warnings
 warnings.filterwarnings('ignore')
 
 # ============================================================
-# CONFIGURATION
+# MONKEY PATCH FOR WHISPER ATTENTION CLASSES (Fix import error)
 # ============================================================
-OUTPUT_DIR = Path("./generated_voiceovers")
-OUTPUT_DIR.mkdir(exist_ok=True)
-
-VOICE_STYLES = {
-    "1": {"name": "Professional Narrator", "emotion": "neutral", "speed": 0.9, "voice": "en-US-JennyNeural"},
-    "2": {"name": "Enthusiastic YouTuber", "emotion": "happy", "speed": 1.1, "voice": "en-US-GuyNeural"},
-    "3": {"name": "Calm Teacher", "emotion": "neutral", "speed": 0.8, "voice": "en-GB-SoniaNeural"},
-    "4": {"name": "Energetic Announcer", "emotion": "happy", "speed": 1.2, "voice": "en-US-DavisNeural"},
-    "5": {"name": "Deep Documentary Voice", "emotion": "neutral", "speed": 0.75, "voice": "en-US-ChristopherNeural"},
-    "6": {"name": "Friendly Explainer", "emotion": "happy", "speed": 1.0, "voice": "en-US-AnaNeural"},
-}
-
-# Templates for voiceover text
-INTROS = [
-    "Watch how", "See how", "Discover how", "Learn how",
-    "Ever wondered how", "Here's how", "This is how",
-    "Let me show you how", "In this video, see how"
-]
-
-ACTION_DESCS = {
-    "increasing torque": ["multiplies rotational force", "boosts pulling power", "enhances mechanical advantage"],
-    "reducing speed": ["controls motion safely", "slows down rotation", "manages velocity"],
-    "transferring power": ["sends energy efficiently", "transmits mechanical force", "delivers power smoothly"],
-    "lifting heavy loads": ["raises massive weights", "hoists heavy objects", "lifts tons with ease"],
-    "driving conveyor belts": ["moves products along", "keeps production flowing", "transports items efficiently"],
-    "precise positioning": ["achieves pinpoint accuracy", "places parts exactly", "aligns with perfection"],
-}
-
-OUTROS = [
-    "That's engineering in action!",
-    "Simple mechanics, powerful results!",
-    "That's how industry gets things done!",
-    "Engineering made simple.",
-    "Powerful mechanics at work.",
-    "Now you know how it works!"
-]
+try:
+    from transformers.models.whisper.modeling_whisper import WHISPER_ATTENTION_CLASSES
+except ImportError:
+    import transformers.models.whisper.modeling_whisper as whisper_module
+    if not hasattr(whisper_module, 'WHISPER_ATTENTION_CLASSES'):
+        whisper_module.WHISPER_ATTENTION_CLASSES = {}
 
 # ============================================================
-# GENERATE VOICEOVER TEXT (Template-based - Always Works)
+# LOAD MINICPM MODEL
 # ============================================================
-def generate_voiceover_text(raw_prompt):
-    """Convert technical prompt to natural voiceover using templates"""
-    
-    if not raw_prompt:
-        return ""
+_model = None
+_tokenizer = None
+
+def load_model():
+    """Load MiniCPM-o-2.6 model for text generation only"""
+    global _model, _tokenizer
     
-    # Parse prompt
-    if " in " in raw_prompt:
-        parts = raw_prompt.split(" in ", 1)
-        mechanism_action = parts[0]
-        industry = parts[1].split(" -")[0] if " -" in parts[1] else parts[1]
-    else:
-        mechanism_action = raw_prompt
-        industry = "industrial setting"
+    if _model is not None:
+        return _model, _tokenizer
     
-    words = mechanism_action.split()
-    mechanism = words[0] if words else "mechanism"
-    action = " ".join(words[1:]) if len(words) > 1 else "operates"
+    print("=" * 60)
+    print("🔄 Loading MiniCPM-o-2.6 model...")
+    print("=" * 60)
     
-    # Find action description
-    import random
-    action_desc = None
-    for key, descs in ACTION_DESCS.items():
-        if key in action.lower():
-            action_desc = random.choice(descs)
-            break
+    from transformers import AutoModel, AutoTokenizer
     
-    if not action_desc:
-        action_desc = f"{action} with precision and reliability"
+    try:
+        _model = AutoModel.from_pretrained(
+            'openbmb/MiniCPM-o-2_6',
+            trust_remote_code=True,
+            torch_dtype=torch.float32,
+            low_cpu_mem_usage=True,
+        )
+    except Exception as e:
+        print(f"⚠️ First attempt error: {e}")
+        print("🔄 Retrying with dtype='auto'...")
+        _model = AutoModel.from_pretrained(
+            'openbmb/MiniCPM-o-2_6',
+            trust_remote_code=True,
+            torch_dtype='auto',
+            low_cpu_mem_usage=True,
+        )
     
-    intro = random.choice(INTROS)
-    outro = random.choice(OUTROS)
-    article = "an" if mechanism[0].lower() in 'aeiou' else "a"
+    _model = _model.eval()
+    _model = _model.to('cpu')
     
-    voiceover = f"{intro} {article} {mechanism} {action} inside {article} {industry}. "
-    voiceover += f"This clever mechanism {action_desc}. {outro}"
+    _tokenizer = AutoTokenizer.from_pretrained(
+        'openbmb/MiniCPM-o-2_6',
+        trust_remote_code=True
+    )
     
-    return voiceover
+    print("✅ MiniCPM-o-2.6 loaded successfully on CPU!")
+    print("=" * 60)
+    return _model, _tokenizer
 
 # ============================================================
-# TEXT TO SPEECH (Edge TTS - Always Works)
+# GENERATE NATURAL VOICEOVER TEXT
 # ============================================================
-def text_to_speech_edge(text, voice_choice, output_path):
-    """Convert text to speech using Microsoft Edge TTS"""
+def generate_natural_text(prompt_text):
+    """Convert technical prompt to natural voiceover text using MiniCPM"""
     
-    import asyncio
-    
-    # Check if edge-tts is installed
-    try:
-        import edge_tts
-    except ImportError:
-        print("📦 Installing edge-tts...")
-        subprocess.check_call([sys.executable, "-m", "pip", "install", "edge-tts"])
-        import edge_tts
-    
-    style = VOICE_STYLES.get(voice_choice, VOICE_STYLES["1"])
-    voice = style["voice"]
+    if not prompt_text:
+        return "⚠️ Please enter a prompt!"
     
-    async def generate():
-        communicate = edge_tts.Communicate(text, voice)
-        await communicate.save(output_path)
+    model, tokenizer = load_model()
     
-    asyncio.run(generate())
-    return output_path
+    # System prompt
+    system_prompt = """You are a professional YouTube Shorts script writer. 
+Convert the given technical engineering prompt into a natural, engaging, and educational voiceover script.
+The script should be 30-50 words long, exciting, easy to speak, and suitable for a short video.
+Return ONLY the script text, no explanations, no quotes."""
 
-# ============================================================
-# MAIN PROCESS FUNCTION
-# ============================================================
-def process_prompt(prompt_text, voice_choice, auto_convert):
-    """Generate voiceover from prompt"""
+    user_message = f"Technical prompt: {prompt_text}\n\nNatural voiceover script:"
     
-    if not prompt_text:
-        return None, "⚠️ Please enter a prompt!", ""
+    msgs = [
+        {'role': 'system', 'content': system_prompt},
+        {'role': 'user', 'content': user_message}
+    ]
     
     try:
-        # Step 1: Generate natural voiceover text
-        if auto_convert:
-            voiceover_text = generate_voiceover_text(prompt_text)
-        else:
-            voiceover_text = prompt_text
-        
-        # Step 2: Generate audio
-        with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp:
-            audio_path = text_to_speech_edge(voiceover_text, voice_choice, tmp.name)
+        response = model.chat(
+            msgs=msgs,
+            tokenizer=tokenizer,
+            sampling=True,
+            temperature=0.7,
+            max_new_tokens=200,
+        )
         
-        # Info message
-        word_count = len(voiceover_text.split())
-        estimated_duration = word_count / 2.5
+        text = response.strip()
         
-        info = f"""
-✅ VOICEOVER GENERATED SUCCESSFULLY!
-
-📝 Original Prompt: {prompt_text}
-
-🎙️ Voiceover Script:
-{voiceover_text}
-
-🎭 Voice Style: {VOICE_STYLES[voice_choice]['name']}
-📊 Word Count: {word_count}
-⏱️ Duration: ~{estimated_duration:.1f} seconds
-        """
+        # Clean up
+        if text.startswith('"'):
+            text = text[1:]
+        if text.endswith('"'):
+            text = text[:-1]
+        if text.lower().startswith('voiceover script:'):
+            text = text[16:].strip()
         
-        return audio_path, info, voiceover_text
+        return text
         
     except Exception as e:
-        return None, f"❌ Error: {str(e)}", ""
+        print(f"❌ Generation error: {e}")
+        return f"Watch how {prompt_text}. That's engineering in action!"
 
 # ============================================================
 # BATCH PROCESS FUNCTION
 # ============================================================
-def batch_process(csv_file, voice_choice, auto_convert):
-    """Process multiple prompts from CSV"""
+def batch_process(file_obj, progress=gr.Progress()):
+    """Process multiple prompts from uploaded CSV"""
     
-    if csv_file is None:
-        return None, "⚠️ Please upload a CSV file!", ""
-    
-    import csv
-    import io
+    if file_obj is None:
+        return "⚠️ Please upload a CSV file!", None
     
     try:
-        # Read CSV
-        content = csv_file.decode('utf-8')
-        csv_reader = csv.DictReader(io.StringIO(content))
+        # Read uploaded CSV
+        content = file_obj.decode('utf-8')
+        lines = content.strip().split('\n')
+        
+        # Parse prompts
+        prompts = []
+        for line in lines[1:]:  # Skip header
+            if line.strip():
+                parts = line.split(',')
+                if len(parts) >= 2:
+                    prompts.append(parts[1].strip().strip('"'))
+                elif len(parts) >= 1:
+                    prompts.append(parts[0].strip().strip('"'))
         
-        batch_dir = OUTPUT_DIR / "batch"
-        batch_dir.mkdir(exist_ok=True)
+        if not prompts:
+            return "❌ No prompts found in CSV!", None
         
+        # Process each prompt
         results = []
-        success_count = 0
+        progress(0, desc="Starting...")
         
-        for idx, row in enumerate(csv_reader, 1):
-            prompt_text = row.get('prompt', row.get('prompt_text', row.get('text', '')))
-            if not prompt_text:
-                continue
-            
-            # Generate voiceover text
-            if auto_convert:
-                voiceover_text = generate_voiceover_text(prompt_text)
-            else:
-                voiceover_text = prompt_text
-            
-            # Generate audio
-            output_path = batch_dir / f"prompt_{idx}.mp3"
-            text_to_speech_edge(voiceover_text, voice_choice, str(output_path))
-            
-            results.append(f"{idx}. {prompt_text[:50]}... → {output_path.name}")
-            success_count += 1
+        for idx, prompt in enumerate(prompts):
+            progress(idx / len(prompts), desc=f"Processing {idx+1}/{len(prompts)}")
+            natural_text = generate_natural_text(prompt)
+            results.append({
+                'prompt': prompt,
+                'voiceover': natural_text,
+                'words': len(natural_text.split())
+            })
+        
+        # Create output CSV
+        output_file = tempfile.NamedTemporaryFile(delete=False, suffix='.csv', mode='w', newline='', encoding='utf-8')
+        writer = csv.DictWriter(output_file, fieldnames=['prompt', 'voiceover', 'words'])
+        writer.writeheader()
+        writer.writerows(results)
+        output_file.close()
         
-        info = f"""
+        # Create summary
+        summary = f"""
 ✅ BATCH PROCESSING COMPLETE!
 
-📊 Total Processed: {success_count}
-🎭 Voice Style: {VOICE_STYLES[voice_choice]['name']}
-📁 Output Folder: {batch_dir}
+📊 Total Prompts: {len(prompts)}
+📝 Successfully Processed: {len(results)}
 
-📋 Generated Files:
-{chr(10).join(results[:10])}
+📋 Sample Output:
+{'-' * 50}
+{results[0]['prompt'][:60]}...
+→ {results[0]['voiceover'][:100]}...
+{'-' * 50}
         """
         
-        return str(batch_dir), info, ""
+        return summary, output_file.name
         
     except Exception as e:
-        return None, f"❌ Error: {str(e)}", ""
+        return f"❌ Error: {str(e)}", None
 
 # ============================================================
-# GRADIO INTERFACE
+# CREATE GRADIO INTERFACE
 # ============================================================
 def create_interface():
-    with gr.Blocks(title="Voiceover Generator", theme=gr.themes.Soft()) as demo:
+    with gr.Blocks(title="MiniCPM Voiceover Text Generator", theme=gr.themes.Soft()) as demo:
         gr.Markdown("""
-        # 🎙️ AI Voiceover Generator
+        # 🎙️ MiniCPM-o-2.6 Voiceover Text Generator
+        
+        **Convert technical prompts to natural voiceover scripts using MiniCPM**
+        
+        | Feature | Description |
+        |---------|-------------|
+        | ✍️ Text Generation | Technical prompt → Natural voiceover script |
+        | 🎯 Pure MiniCPM | No fallbacks, only MiniCPM-o-2.6 |
+        | ⚡ CPU Only | Works on your 16GB RAM / 2 vCPU |
+        | 📁 Batch Processing | Upload CSV and process multiple prompts |
         
-        **Turn technical prompts into natural voiceovers - No setup required!**
+        ---
         
-        Just type a prompt like `"planetary gear increasing torque in automotive plant"` and click generate.
+        **Example:** `"planetary gear increasing torque in automotive plant"`
         """)
         
         with gr.Tabs():
@@ -242,49 +216,65 @@ def create_interface():
                         prompt_input = gr.Textbox(
                             label="📝 Technical Prompt",
                             placeholder="Example: planetary gear increasing torque in automotive plant",
-                            lines=3
+                            lines=4
                         )
                         
                         with gr.Row():
-                            voice_dropdown = gr.Dropdown(
-                                choices=list(VOICE_STYLES.keys()),
-                                label="🎭 Voice Style",
-                                value="1",
-                                info="1=Professional, 2=YouTuber, 3=Teacher, 4=Announcer, 5=Documentary, 6=Friendly"
-                            )
-                            
-                            auto_convert = gr.Checkbox(
-                                label="✨ Auto-convert to natural voiceover",
-                                value=True,
-                                info="Convert technical prompt to engaging script"
-                            )
-                        
-                        with gr.Row():
-                            generate_btn = gr.Button("🎤 Generate Voiceover", variant="primary", size="lg")
+                            generate_btn = gr.Button("🎤 Generate Voiceover Text", variant="primary", size="lg")
                             clear_btn = gr.Button("🗑️ Clear", size="lg")
+                        
+                        gr.Markdown("""
+                        ### 💡 Example Prompts
+                        
+                        Click any to try:
+                        """)
+                        
+                        example_prompts = [
+                            "planetary gear increasing torque in automotive plant",
+                            "worm gear reducing speed in steel mill",
+                            "hydraulic cylinder lifting heavy loads in mining operation",
+                            "chain drive driving conveyor belts in food processing factory",
+                            "servo motor precise positioning in warehouse",
+                            "ball bearing reducing friction in high speed machinery",
+                            "rack and pinion converting rotary to linear motion",
+                            "camshaft opening and closing valves in engine"
+                        ]
+                        
+                        for ex in example_prompts:
+                            gr.Markdown(f"- `{ex}`")
                     
                     with gr.Column(scale=1):
                         gr.Markdown("""
-                        ### 🎭 Voice Style Guide
+                        ### 🎯 What MiniCPM Does
+                        
+                        1. **Understands** your technical prompt
+                        2. **Creates** natural, engaging voiceover script
+                        3. **Optimizes** for YouTube Shorts (30-50 words)
+                        4. **Returns** ready-to-use script
                         
-                        | # | Style | Best For |
-                        |---|-------|----------|
-                        | 1 | Professional | Technical explainers |
-                        | 2 | YouTuber | Engaging shorts |
-                        | 3 | Teacher | Educational |
-                        | 4 | Announcer | High energy |
-                        | 5 | Documentary | Serious topics |
-                        | 6 | Friendly | Casual tutorials |
+                        ### 📊 Output Format
+                        
+                        - Natural, conversational tone
+                        - Engaging hooks
+                        - Clear explanations
+                        - Strong conclusions
                         """)
                 
                 with gr.Row():
                     with gr.Column():
-                        audio_output = gr.Audio(label="🎧 Generated Voiceover", type="filepath")
-                        download_btn = gr.File(label="📥 Download Audio", visible=False)
-                    
-                    with gr.Column():
-                        info_output = gr.Textbox(label="📊 Status", lines=6)
-                        text_output = gr.Textbox(label="📝 Voiceover Script", lines=4)
+                        output_text = gr.Textbox(
+                            label="🎙️ Generated Voiceover Script",
+                            lines=6,
+                            interactive=False
+                        )
+                        
+                        info_text = gr.Textbox(
+                            label="📊 Info",
+                            lines=3,
+                            interactive=False
+                        )
+                        
+                        copy_btn = gr.Button("📋 Copy to Clipboard", size="sm")
             
             # Tab 2: Batch Processing
             with gr.TabItem("📁 Batch Processing"):
@@ -294,90 +284,114 @@ def create_interface():
                             label="Upload CSV File",
                             file_types=[".csv"],
                             type="binary",
-                            info="CSV should have a 'prompt' column"
-                        )
-                        
-                        batch_voice = gr.Dropdown(
-                            choices=list(VOICE_STYLES.keys()),
-                            label="🎭 Voice Style",
-                            value="1"
-                        )
-                        
-                        batch_auto_convert = gr.Checkbox(
-                            label="✨ Auto-convert prompts",
-                            value=True
-                        )
-                        
-                        batch_max_rows = gr.Slider(
-                            label="Maximum prompts (0 = all)",
-                            minimum=0,
-                            maximum=100,
-                            value=10,
-                            step=1
+                            info="CSV should have header: id,prompt or just prompts column"
                         )
                         
                         batch_btn = gr.Button("🚀 Start Batch Processing", variant="primary")
                     
                     with gr.Column():
-                        batch_output = gr.Textbox(label="📁 Output Folder", lines=2)
-                        batch_info = gr.Textbox(label="📊 Batch Status", lines=15)
+                        batch_status = gr.Textbox(label="📊 Status", lines=10)
+                        batch_output_file = gr.File(label="📥 Download Results", visible=True)
             
-            # Tab 3: Examples
-            with gr.TabItem("📋 Examples"):
+            # Tab 3: Help
+            with gr.TabItem("❓ Help"):
                 gr.Markdown("""
-                ## 📝 Try These Prompts
+                ## 📖 How to Use
+                
+                ### Single Prompt Mode
+                1. Enter a technical prompt in the text box
+                2. Click "Generate Voiceover Text"
+                3. Copy the generated script for your video
                 
-                Copy any prompt and paste in the Single Prompt tab:
+                ### Batch Mode
+                1. Prepare a CSV file with this format:
+                   ```csv
+                   id,prompt
+                   1,planetary gear increasing torque in automotive plant
+                   2,worm gear reducing speed in steel mill
+                   ```
+                2. Upload the CSV file
+                3. Click "Start Batch Processing"
+                4. Download results CSV
                 
+                ## 🎯 Prompt Format
+                
+                Best results with this format:
                 ```
-                planetary gear increasing torque in automotive plant
-                worm gear reducing speed in steel mill
-                hydraulic cylinder lifting heavy loads in mining operation
-                chain drive driving conveyor belts in food processing factory
-                servo motor precise positioning in warehouse
-                ball bearing reducing friction in high speed machinery
-                rack and pinion converting rotary to linear motion
-                camshaft opening and closing valves in engine
-                bevel gear transferring power at right angle
-                clutch assembly engaging and disengaging power
+                [mechanism] [action] in [industry]
                 ```
                 
-                ## 📁 CSV Format for Batch Processing
+                Examples:
+                - `planetary gear increasing torque in automotive plant`
+                - `worm gear reducing speed in steel mill`
+                - `hydraulic cylinder lifting heavy loads in mining`
                 
-                Create a CSV file with this structure:
+                ## ⚙️ Technical Details
                 
-                ```csv
-                id,prompt
-                1,planetary gear increasing torque in automotive plant
-                2,worm gear reducing speed in steel mill
-                3,hydraulic cylinder lifting heavy loads in mining operation
-                ```
+                - **Model:** MiniCPM-o-2.6 (8B parameters)
+                - **Hardware:** CPU only (16GB RAM recommended)
+                - **First run:** Downloads model (~8GB, 5-10 minutes)
+                - **Subsequent runs:** Instant loading from cache
                 
-                ## 💡 Tips
+                ## 🔧 Troubleshooting
                 
-                - **Auto-convert ON:** Technical prompt automatically becomes natural voiceover
-                - **Auto-convert OFF:** Use your exact text as voiceover
-                - First generation may take 2-3 seconds (installing dependencies)
-                - Audio downloads automatically in MP3 format
+                | Issue | Solution |
+                |-------|----------|
+                | Slow first generation | Model downloading in background |
+                | Out of memory | Close other applications |
+                | Import errors | Use Docker for consistent environment |
                 """)
         
         # Event handlers
+        def process_and_info(prompt):
+            if not prompt:
+                return "⚠️ Please enter a prompt!", "❌ No prompt provided"
+            
+            result = generate_natural_text(prompt)
+            word_count = len(result.split())
+            info = f"""
+✅ Generated successfully!
+
+📝 Original prompt: {prompt[:60]}...
+📊 Word count: {word_count}
+⏱️ Estimated duration: ~{word_count / 2.5:.1f} seconds
+🤖 Model: MiniCPM-o-2.6
+            """
+            return result, info
+        
         generate_btn.click(
-            fn=process_prompt,
-            inputs=[prompt_input, voice_dropdown, auto_convert],
-            outputs=[audio_output, info_output, text_output]
+            fn=process_and_info,
+            inputs=[prompt_input],
+            outputs=[output_text, info_text]
         )
         
         clear_btn.click(
             fn=lambda: ("", "", ""),
-            outputs=[prompt_input, info_output, text_output]
+            outputs=[prompt_input, output_text, info_text]
+        )
+        
+        copy_btn.click(
+            fn=lambda: output_text.value,
+            outputs=None,
+            js="(text) => navigator.clipboard.writeText(text)"
         )
         
         batch_btn.click(
             fn=batch_process,
-            inputs=[csv_input, batch_voice, batch_auto_convert],
-            outputs=[batch_output, batch_info, batch_info]
+            inputs=[csv_input],
+            outputs=[batch_status, batch_output_file]
         )
+        
+        # Example prompt click (using JavaScript)
+        example_js = """
+        function set_example(prompt) {
+            document.querySelector('textarea').value = prompt;
+            return prompt;
+        }
+        """
+        
+        for ex in example_prompts[:4]:
+            gr.Markdown(f"<button onclick='set_example(\"{ex}\")'>{ex}</button>", elem_id="example-btn")
     
     return demo
 
@@ -386,14 +400,9 @@ def create_interface():
 # ============================================================
 if __name__ == "__main__":
     print("=" * 60)
-    print("🎙️ AI VOICEOVER GENERATOR")
+    print("🎙️ MiniCPM-o-2.6 Voiceover Text Generator")
     print("=" * 60)
-    print("\n✨ Features:")
-    print("   - Natural voiceover text from technical prompts")
-    print("   - 6 professional voice styles")
-    print("   - Batch processing from CSV")
-    print("   - No GPU required (uses Edge TTS)")
-    print("\n🚀 Starting application...")
+    print("🚀 Starting Gradio Interface...")
     print("📱 Access at: http://localhost:7860")
     print("=" * 60)