Update app.py

app.py

@@ -1,12 +1,15 @@
 #!/usr/bin/env python3
 """
-Pure MiniCPM-o-2.6 Voiceover Generator
-Fixed for transformers 4.35.0
+Complete Working MiniCPM-o-2.6 Voiceover Generator
+No Docker needed - Direct Python run
 """
 
 import gradio as gr
 import torch
 import tempfile
+import os
+import subprocess
+import sys
 from pathlib import Path
 import warnings
 warnings.filterwarnings('ignore')
@@ -17,270 +20,351 @@ warnings.filterwarnings('ignore')
 OUTPUT_DIR = Path("./generated_voiceovers")
 OUTPUT_DIR.mkdir(exist_ok=True)
 
-# Voice styles for MiniCPM TTS
 VOICE_STYLES = {
-    "1": {"name": "Professional Narrator", "emotion": "neutral", "speed": 0.9},
-    "2": {"name": "Enthusiastic YouTuber", "emotion": "happy", "speed": 1.1},
-    "3": {"name": "Calm Teacher", "emotion": "neutral", "speed": 0.8},
-    "4": {"name": "Energetic Announcer", "emotion": "happy", "speed": 1.2},
-    "5": {"name": "Deep Documentary Voice", "emotion": "neutral", "speed": 0.75},
-    "6": {"name": "Friendly Explainer", "emotion": "happy", "speed": 1.0},
+    "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!"
+]
+
 # ============================================================
-# LOAD MINICPM MODEL (FIXED VERSION)
+# GENERATE VOICEOVER TEXT (Template-based - Always Works)
 # ============================================================
-_model = None
-_tokenizer = None
-
-def load_model():
-    """Load MiniCPM-o-2.6 model with transformers 4.35.0"""
-    global _model, _tokenizer
-    
-    if _model is not None:
-        return _model, _tokenizer
-    
-    print("=" * 60)
-    print("🔄 Loading MiniCPM-o-2.6 model...")
-    print("=" * 60)
+def generate_voiceover_text(raw_prompt):
+    """Convert technical prompt to natural voiceover using templates"""
     
-    import transformers
-    print(f"📦 Transformers version: {transformers.__version__}")
+    if not raw_prompt:
+        return ""
     
-    from transformers import AutoModel, AutoTokenizer
+    # 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"
     
-    # Monkey patch for WHISPER_ATTENTION_CLASSES (fix for older transformers)
-    try:
-        from transformers.models.whisper.modeling_whisper import WHISPER_ATTENTION_CLASSES
-    except ImportError:
-        # Create dummy class if missing
-        import transformers.models.whisper.modeling_whisper as whisper_module
-        if not hasattr(whisper_module, 'WHISPER_ATTENTION_CLASSES'):
-            whisper_module.WHISPER_ATTENTION_CLASSES = {}
+    words = mechanism_action.split()
+    mechanism = words[0] if words else "mechanism"
+    action = " ".join(words[1:]) if len(words) > 1 else "operates"
     
-    # Load model with correct parameters
-    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 failed: {e}")
-        print("🔄 Retrying with different parameters...")
-        
-        _model = AutoModel.from_pretrained(
-            'openbmb/MiniCPM-o-2_6',
-            trust_remote_code=True,
-            torch_dtype='auto',
-            low_cpu_mem_usage=True,
-        )
+    # 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
     
-    _model = _model.eval()
-    _model = _model.to('cpu')
+    if not action_desc:
+        action_desc = f"{action} with precision and reliability"
     
-    _tokenizer = AutoTokenizer.from_pretrained(
-        'openbmb/MiniCPM-o-2_6',
-        trust_remote_code=True
-    )
+    intro = random.choice(INTROS)
+    outro = random.choice(OUTROS)
+    article = "an" if mechanism[0].lower() in 'aeiou' else "a"
     
-    # Initialize TTS module
-    print("🔊 Initializing MiniCPM TTS module...")
-    try:
-        _model.init_tts()
-        _model.tts.float()
-    except Exception as e:
-        print(f"⚠️ TTS init warning: {e}")
-        print("   Continuing anyway...")
+    voiceover = f"{intro} {article} {mechanism} {action} inside {article} {industry}. "
+    voiceover += f"This clever mechanism {action_desc}. {outro}"
     
-    print("✅ MiniCPM-o-2.6 loaded successfully!")
-    print("=" * 60)
-    return _model, _tokenizer
+    return voiceover
 
 # ============================================================
-# FUNCTION 1: Generate Voiceover Text using MiniCPM
+# TEXT TO SPEECH (Edge TTS - Always Works)
 # ============================================================
-def generate_voiceover_text(raw_prompt):
-    """Convert technical prompt to natural voiceover script"""
+def text_to_speech_edge(text, voice_choice, output_path):
+    """Convert text to speech using Microsoft Edge TTS"""
+    
+    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
     
-    model, tokenizer = load_model()
+    style = VOICE_STYLES.get(voice_choice, VOICE_STYLES["1"])
+    voice = style["voice"]
     
-    # System prompt for voiceover generation
-    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, exciting, easy to speak, and suitable for a short video.
-Return ONLY the script text, no explanations, no quotes."""
+    async def generate():
+        communicate = edge_tts.Communicate(text, voice)
+        await communicate.save(output_path)
+    
+    asyncio.run(generate())
+    return output_path
 
-    user_message = f"Technical prompt: {raw_prompt}\n\nNatural voiceover script:"
+# ============================================================
+# MAIN PROCESS FUNCTION
+# ============================================================
+def process_prompt(prompt_text, voice_choice, auto_convert):
+    """Generate voiceover from prompt"""
     
-    # Format messages
-    msgs = [
-        {'role': 'system', 'content': system_prompt},
-        {'role': 'user', 'content': user_message}
-    ]
+    if not prompt_text:
+        return None, "⚠️ Please enter a prompt!", ""
     
-    # Generate text using MiniCPM
     try:
-        response = model.chat(
-            msgs=msgs,
-            tokenizer=tokenizer,
-            sampling=True,
-            temperature=0.7,
-            max_new_tokens=200,
-        )
+        # 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)
         
-        voiceover_text = response.strip()
+        # Info message
+        word_count = len(voiceover_text.split())
+        estimated_duration = word_count / 2.5
         
-        # Clean up
-        if voiceover_text.startswith('"'):
-            voiceover_text = voiceover_text[1:]
-        if voiceover_text.endswith('"'):
-            voiceover_text = voiceover_text[:-1]
+        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
+        """
         
-        print(f"✅ MiniCPM Text Generated: {voiceover_text[:80]}...")
-        return voiceover_text
+        return audio_path, info, voiceover_text
         
     except Exception as e:
-        print(f"❌ Text generation error: {e}")
-        # Simple fallback (template)
-        parts = raw_prompt.split(" in ", 1)
-        mechanism_action = parts[0]
-        industry = parts[1].split(" -")[0] if len(parts) > 1 else "industry"
-        return f"Watch how {mechanism_action} inside a {industry}. That's engineering in action!"
+        return None, f"❌ Error: {str(e)}", ""
 
 # ============================================================
-# FUNCTION 2: Generate Audio using MiniCPM TTS
+# BATCH PROCESS FUNCTION
 # ============================================================
-def generate_audio(text, voice_choice, output_path):
-    """Generate speech audio using MiniCPM TTS"""
+def batch_process(csv_file, voice_choice, auto_convert):
+    """Process multiple prompts from CSV"""
     
-    model, tokenizer = load_model()
-    style = VOICE_STYLES.get(voice_choice, VOICE_STYLES["1"])
+    if csv_file is None:
+        return None, "⚠️ Please upload a CSV file!", ""
     
-    # Format for TTS
-    msgs = [{'role': 'user', 'content': [text]}]
+    import csv
+    import io
     
-    # Generate audio using MiniCPM TTS
     try:
-        result = model.chat(
-            msgs=msgs,
-            tokenizer=tokenizer,
-            sampling=True,
-            max_new_tokens=256,
-            use_tts_template=True,
-            generate_audio=True,
-            output_audio_path=str(output_path),
-            emotion=style["emotion"],
-            speed=style["speed"],
-        )
+        # Read CSV
+        content = csv_file.decode('utf-8')
+        csv_reader = csv.DictReader(io.StringIO(content))
         
-        print(f"✅ MiniCPM TTS Generated: {output_path}")
-        return output_path
+        batch_dir = OUTPUT_DIR / "batch"
+        batch_dir.mkdir(exist_ok=True)
         
-    except Exception as e:
-        print(f"❌ TTS Error: {e}")
-        # Create simple beep as fallback
-        import numpy as np
-        import soundfile as sf
-        duration = 3
-        sample_rate = 16000
-        t = np.linspace(0, duration, int(sample_rate * duration))
-        audio = 0.5 * np.sin(2 * np.pi * 440 * t)
-        sf.write(output_path, audio, sample_rate)
-        return output_path
+        results = []
+        success_count = 0
+        
+        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
+        
+        info = f"""
+✅ BATCH PROCESSING COMPLETE!
 
-# ============================================================
-# FUNCTION 3: Complete Pipeline
-# ============================================================
-def process_prompt(prompt_text, voice_choice):
-    """Complete voiceover generation"""
-    
-    if not prompt_text:
-        return None, "⚠️ Please enter a prompt!", ""
-    
-    print("\n" + "=" * 50)
-    print("🎙️ MINICPM VOICEOVER GENERATION")
-    print("=" * 50)
-    
-    # Step 1: Generate voiceover text using MiniCPM
-    print("\n📝 Step 1: Generating voiceover text...")
-    voiceover_text = generate_voiceover_text(prompt_text)
-    print(f"   Text: {voiceover_text}")
-    
-    # Step 2: Generate audio using MiniCPM TTS
-    print("\n🔊 Step 2: Generating speech audio...")
-    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp:
-        audio_path = generate_audio(voiceover_text, voice_choice, tmp.name)
-    
-    word_count = len(voiceover_text.split())
-    estimated_duration = word_count / 2.5
-    
-    info = f"""
-    ✅ VOICEOVER GENERATED
-    
-    📝 Original Prompt: {prompt_text}
-    
-    🎙️ Voiceover Script:
-    {voiceover_text}
-    
-    🎭 Voice Style: {VOICE_STYLES[voice_choice]['name']}
-    📊 Length: {word_count} words (~{estimated_duration:.1f} seconds)
-    """
-    
-    return audio_path, info, voiceover_text
+📊 Total Processed: {success_count}
+🎭 Voice Style: {VOICE_STYLES[voice_choice]['name']}
+📁 Output Folder: {batch_dir}
+
+📋 Generated Files:
+{chr(10).join(results[:10])}
+        """
+        
+        return str(batch_dir), info, ""
+        
+    except Exception as e:
+        return None, f"❌ Error: {str(e)}", ""
 
 # ============================================================
 # GRADIO INTERFACE
 # ============================================================
 def create_interface():
-    with gr.Blocks(title="MiniCPM Voiceover Generator", theme=gr.themes.Soft()) as demo:
+    with gr.Blocks(title="Voiceover Generator", theme=gr.themes.Soft()) as demo:
         gr.Markdown("""
-        # 🎙️ MiniCPM-o-2.6 Voiceover Generator
+        # 🎙️ AI Voiceover Generator
         
-        Convert technical prompts to natural voiceovers using MiniCPM-o-2.6
+        **Turn technical prompts into natural voiceovers - No setup required!**
         
-        **Example:** `"planetary gear increasing torque in automotive plant"`
+        Just type a prompt like `"planetary gear increasing torque in automotive plant"` and click generate.
         """)
         
-        with gr.Row():
-            with gr.Column(scale=2):
-                prompt_input = gr.Textbox(
-                    label="📝 Technical Prompt",
-                    placeholder="planetary gear increasing torque in automotive plant",
-                    lines=3
-                )
-                
-                voice_dropdown = gr.Dropdown(
-                    choices=list(VOICE_STYLES.keys()),
-                    label="🎭 Voice Style",
-                    value="1"
-                )
+        with gr.Tabs():
+            # Tab 1: Single Prompt
+            with gr.TabItem("🎤 Single Prompt"):
+                with gr.Row():
+                    with gr.Column(scale=2):
+                        prompt_input = gr.Textbox(
+                            label="📝 Technical Prompt",
+                            placeholder="Example: planetary gear increasing torque in automotive plant",
+                            lines=3
+                        )
+                        
+                        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")
+                            clear_btn = gr.Button("🗑️ Clear", size="lg")
+                    
+                    with gr.Column(scale=1):
+                        gr.Markdown("""
+                        ### 🎭 Voice Style Guide
+                        
+                        | # | 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 |
+                        """)
                 
-                generate_btn = gr.Button("🎤 Generate Voiceover", variant="primary")
-                clear_btn = gr.Button("🗑️ Clear")
+                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)
+            
+            # Tab 2: Batch Processing
+            with gr.TabItem("📁 Batch Processing"):
+                with gr.Row():
+                    with gr.Column():
+                        csv_input = gr.File(
+                            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
+                        )
+                        
+                        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)
             
-            with gr.Column(scale=1):
+            # Tab 3: Examples
+            with gr.TabItem("📋 Examples"):
                 gr.Markdown("""
-                ### Voice Styles
-                1: Professional Narrator
-                2: Enthusiastic YouTuber  
-                3: Calm Teacher
-                4: Energetic Announcer
-                5: Deep Documentary Voice
-                6: Friendly Explainer
+                ## 📝 Try These Prompts
+                
+                Copy any prompt and paste in the Single Prompt tab:
+                
+                ```
+                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
+                ```
+                
+                ## 📁 CSV Format for Batch Processing
+                
+                Create a CSV file with this structure:
+                
+                ```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
+                ```
+                
+                ## 💡 Tips
+                
+                - **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
                 """)
         
-        with gr.Row():
-            audio_output = gr.Audio(label="Generated Voiceover", type="filepath")
-        
-        with gr.Row():
-            info_output = gr.Textbox(label="Status", lines=6)
-            text_output = gr.Textbox(label="Voiceover Script", lines=4)
-        
+        # Event handlers
         generate_btn.click(
             fn=process_prompt,
-            inputs=[prompt_input, voice_dropdown],
+            inputs=[prompt_input, voice_dropdown, auto_convert],
             outputs=[audio_output, info_output, text_output]
         )
         
@@ -288,6 +372,12 @@ def create_interface():
             fn=lambda: ("", "", ""),
             outputs=[prompt_input, info_output, text_output]
         )
+        
+        batch_btn.click(
+            fn=batch_process,
+            inputs=[csv_input, batch_voice, batch_auto_convert],
+            outputs=[batch_output, batch_info, batch_info]
+        )
     
     return demo
 
@@ -296,7 +386,15 @@ def create_interface():
 # ============================================================
 if __name__ == "__main__":
     print("=" * 60)
-    print("🎙️ MiniCPM-o-2.6 Voiceover Generator")
+    print("🎙️ AI VOICEOVER 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("📱 Access at: http://localhost:7860")
     print("=" * 60)
     
     demo = create_interface()