app.py

#!/usr/bin/env python3
"""
MiloMusic - Hugging Face Spaces Version
AI-powered music generation platform optimized for cloud deployment with high-performance configuration.
"""

import multiprocessing
import os
import sys
import subprocess
import tempfile
import gradio as gr
import soundfile as sf
from dataclasses import dataclass, field
from typing import Any
import xxhash
import numpy as np
import spaces
import groq

# Import environment setup for Spaces
def setup_spaces_environment():
    """Setup environment variables and paths for Hugging Face Spaces"""
    # Set HuggingFace cache directory
    os.environ["HF_HOME"] = "/tmp/hf_cache"
    os.environ["TRANSFORMERS_CACHE"] = "/tmp/transformers_cache"
    os.environ["HF_HUB_CACHE"] = "/tmp/hf_hub_cache"
    
    # 1.PyTorch CUDA memory optimization 2.用PyTorch的可扩展内存段分配, 提高GPU内存使用效率, 减少内存碎片问题
    os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "expandable_segments:True"
    
    # Set temp directory for audio files
    os.environ["TMPDIR"] = "/tmp"
    
    print("🚀 Environment setup complete for Spaces")

# Install flash-attn if not already installed
def install_flash_attn() -> bool:
    """Install flash-attn from source with proper compilation flags"""
    try:
        import flash_attn
        print("✅ flash-attn already installed")
        return True
    except ImportError:
        print("📦 Installing flash-attn from source...")
        try:
            # Install with optimized settings for Spaces
            cmd = [
                sys.executable, "-m", "pip", "install", 
                "--no-build-isolation", 
                "--no-cache-dir",
                "flash-attn",
                "--verbose"
            ]
            
            # Use more parallel jobs for faster compilation in Spaces
            env = os.environ.copy()
            max_jobs = min(4, multiprocessing.cpu_count())  # Utilize more CPU cores
            env["MAX_JOBS"] = str(max_jobs)
            env["NVCC_PREPEND_FLAGS"] = "-ccbin /usr/bin/gcc"
            
            result = subprocess.run(cmd, env=env, capture_output=True, text=True, timeout=1800)  # 30 min timeout
            
            if result.returncode == 0:
                print("✅ flash-attn installed successfully")
                return True
            else:
                print(f"❌ flash-attn installation failed: {result.stderr}")
                return False
                
        except subprocess.TimeoutExpired:
            print("⏰ flash-attn installation timed out")
            return False
        except Exception as e:
            print(f"❌ Error installing flash-attn: {e}")
            return False

# Setup environment first
setup_spaces_environment()

# Download required models for YuEGP inference
def download_required_models():
    """Download required model files at startup"""
    try:
        from download_models import ensure_model_availability
        print("🚀 Checking and downloading required models...")
        success = ensure_model_availability()
        if success:
            print("✅ Model setup completed successfully")
        else:
            print("⚠️ Some models may be missing - continuing with available resources")
        return success
    except ImportError as e:
        print(f"⚠️ Model download script not found: {e}")
        return False
    except Exception as e:
        print(f"❌ Error during model download: {e}")
        return False

# Download models before other setup
models_ready = download_required_models()

# Install flash-attn if needed
flash_attn_available = install_flash_attn()

# Apply transformers patches for performance optimization
def apply_transformers_patch():
    """
    Apply YuEGP transformers patches for high-performance generation.
    
    This function applies optimized transformers patches that provide:
    - 2x speed improvement for low VRAM profiles
    - 3x speed improvement for Stage 1 generation (16GB+ VRAM)
    - 2x speed improvement for Stage 2 generation (all profiles)
    
    The patches replace two key files in the transformers library:
    - models/llama/modeling_llama.py (LLaMA model optimizations)
    - generation/utils.py (generation utilities optimizations)
    
    Includes smart detection to avoid re-applying patches on restart.
    """
    try:
        import shutil
        import site
        import hashlib
        
        # Define source and target directories
        source_dir = os.path.join(project_root, "YuEGP", "transformers")
        
        # Get the site-packages directory where transformers is installed
        site_packages = site.getsitepackages()
        if not site_packages:
            # Fallback for some environments
            import transformers
            transformers_path = os.path.dirname(transformers.__file__)
            target_base = os.path.dirname(transformers_path)
        else:
            target_base = site_packages[0]
        
        target_dir = os.path.join(target_base, "transformers")
        
        # Check if source patches exist
        if not os.path.exists(source_dir):
            print("⚠️  YuEGP transformers patches not found, skipping optimization")
            return False
        
        if not os.path.exists(target_dir):
            print("⚠️  Transformers library not found, skipping patches")
            return False
        
        # Check if patches are already applied by comparing file hashes
        def get_file_hash(filepath):
            """Get MD5 hash of file content"""
            if not os.path.exists(filepath):
                return None
            with open(filepath, 'rb') as f:
                return hashlib.md5(f.read()).hexdigest()
        
        # Key files to check for patch status
        key_patches = [
            "models/llama/modeling_llama.py",
            "generation/utils.py"
        ]
        
        patches_needed = False
        for patch_file in key_patches:
            source_file = os.path.join(source_dir, patch_file)
            target_file = os.path.join(target_dir, patch_file)
            
            if os.path.exists(source_file):
                source_hash = get_file_hash(source_file)
                target_hash = get_file_hash(target_file)
                
                if source_hash != target_hash:
                    patches_needed = True
                    break
        
        if not patches_needed:
            print("✅ YuEGP transformers patches already applied, skipping re-installation")
            print("  📈 High-performance optimizations are active:")
            print("    • Stage 1 generation: 3x faster (16GB+ VRAM)")
            print("    • Stage 2 generation: 2x faster (all profiles)")
            return True
        
        # Apply patches by copying optimized files
        print("🔧 Applying YuEGP transformers patches for high-performance generation...")
        
        # Copy the patched files, preserving directory structure
        for root, dirs, files in os.walk(source_dir):
            # Calculate relative path from source_dir
            rel_path = os.path.relpath(root, source_dir)
            target_subdir = os.path.join(target_dir, rel_path) if rel_path != '.' else target_dir
            
            # Ensure target subdirectory exists
            os.makedirs(target_subdir, exist_ok=True)
            
            # Copy all Python files in this directory
            for file in files:
                if file.endswith('.py'):
                    src_file = os.path.join(root, file)
                    dst_file = os.path.join(target_subdir, file)
                    
                    shutil.copy2(src_file, dst_file)
                    print(f"  ✅ Patched: {os.path.relpath(dst_file, target_base)}")
        
        print("🚀 Transformers patches applied successfully!")
        print("  📈 Expected performance gains:")
        print("    • Stage 1 generation: 3x faster (16GB+ VRAM)")
        print("    • Stage 2 generation: 2x faster (all profiles)")
        return True
        
    except Exception as e:
        print(f"❌ Error applying transformers patches: {e}")
        print("   Continuing without patches - performance may be reduced")
        return False

# Now import the rest of the dependencies
# Add project root to Python path for imports
project_root = os.path.dirname(os.path.abspath(__file__))
if project_root not in sys.path:
    sys.path.insert(0, project_root)

from tools.groq_client import client as groq_client
from openai import OpenAI
from tools.generate_lyrics import generate_structured_lyrics, format_lyrics

# Apply patches after all imports are set up
patch_applied = apply_transformers_patch()

# Import CUDA info after flash-attn setup
import torch
if torch.cuda.is_available():
    print(f"🎮 GPU: {torch.cuda.get_device_name(0)}")
    print(f"💾 VRAM: {torch.cuda.get_device_properties(0).total_memory / 1e9:.1f}GB")
else:
    print("⚠️  No CUDA GPU detected")

@dataclass
class AppState:
    """
    Maintains the application state throughout user interactions.
    """
    conversation: list = field(default_factory=list)
    stopped: bool = False
    model_outs: Any = None
    lyrics: str = ""
    genre: str = "pop"
    mood: str = "upbeat"
    theme: str = "love"

def validate_api_keys():
    """Validate required API keys for Spaces deployment"""
    required_keys = ["GROQ_API_KEY",  "GEMINI_API_KEY"]
    missing_keys = []
    
    for key in required_keys:
        if not os.getenv(key):
            missing_keys.append(key)
    
    if missing_keys:
        print(f"⚠️  Missing API keys: {missing_keys}")
        return False
    
    print("✅ All API keys validated")
    return True

def validate_file_structure():
    """Validate that required files and directories exist"""
    required_paths = [
        "YuEGP/inference/infer.py",
        "YuEGP/inference/codecmanipulator.py", 
        "YuEGP/inference/mmtokenizer.py",
        "tools/generate_lyrics.py",
        "tools/groq_client.py",
        "schemas/lyrics.py"  # Required for lyrics structure models
    ]
    
    missing_files = []
    for path in required_paths:
        if not os.path.exists(path):
            missing_files.append(path)
    
    if missing_files:
        print(f"⚠️  Missing required files: {missing_files}")
        return False
    
    print("✅ All required files found")
    return True

@spaces.GPU(duration=1200) # H200 on ZeroGPU is free for 25mins, for compatibility on A10G large and L40s
def generate_music_spaces(lyrics: str, genre: str, mood: str, progress=gr.Progress()) -> str:
    """
    Generate music using YuE model with high-performance Spaces configuration
    """
    if not lyrics.strip():
        return "Please provide lyrics to generate music."
    
    try:
        progress(0.1, desc="Preparing lyrics...")
        
        # Use lyrics directly (already formatted from chat interface)
        formatted_lyrics = lyrics
        
        # Create temporary files
        with tempfile.NamedTemporaryFile(mode='w', suffix='.txt', delete=False) as genre_file:
            genre_file.write(f"instrumental,{genre},{mood},male vocals")
            genre_file_path = genre_file.name
        
        # Convert lyrics format for YuEGP compatibility
        # YuEGP expects [VERSE], [CHORUS] format, but our AI generates **VERSE**, **CHORUS**
        import re
        
        # Extract only the actual lyrics content, removing AI commentary
        formatted_lyrics_for_yue = formatted_lyrics
        
        # Convert **VERSE 1** to [VERSE], **CHORUS** to [CHORUS], etc.
        formatted_lyrics_for_yue = re.sub(r'\*\*(VERSE\s*\d*)\*\*', r'[\1]', formatted_lyrics_for_yue)
        formatted_lyrics_for_yue = re.sub(r'\*\*(CHORUS)\*\*', r'[\1]', formatted_lyrics_for_yue)
        formatted_lyrics_for_yue = re.sub(r'\*\*(BRIDGE)\*\*', r'[\1]', formatted_lyrics_for_yue)
        formatted_lyrics_for_yue = re.sub(r'\*\*(OUTRO)\*\*', r'[\1]', formatted_lyrics_for_yue)
        
        # Remove AI commentary (lines that don't contain actual lyrics)
        lines = formatted_lyrics_for_yue.split('\n')
        clean_lines = []
        in_song = False
        
        for line in lines:
            line = line.strip()
            # Start collecting from first section marker
            if re.match(r'\[(VERSE|CHORUS|BRIDGE|OUTRO)', line):
                in_song = True
            # Stop at AI commentary
            if in_song and line and not line.startswith('[') and any(phrase in line.lower() for phrase in ['how do you like', 'would you like', 'let me know', 'take a look']):
                break
            if in_song:
                clean_lines.append(line)
        
        formatted_lyrics_for_yue = '\n'.join(clean_lines).strip()
        
        print(f"🐛 DEBUG - Original lyrics length: {len(formatted_lyrics)}")
        print(f"🐛 DEBUG - Converted lyrics for YuE: '{formatted_lyrics_for_yue}'")
        print(f"🐛 DEBUG - Converted lyrics length: {len(formatted_lyrics_for_yue)}")
        
        with tempfile.NamedTemporaryFile(mode='w', suffix='.txt', delete=False) as lyrics_file:
            lyrics_file.write(formatted_lyrics_for_yue)
            lyrics_file_path = lyrics_file.name
        
        progress(0.2, desc="Setting up generation...")
        
        # Generate music with high-performance Spaces configuration
        output_dir = tempfile.mkdtemp()
        
        # High-performance command based on Spaces GPU resources
        # In Spaces, working directory is /app
        infer_script_path = os.path.join(os.getcwd(), "YuEGP", "inference", "infer.py")
        cmd = [
            sys.executable,
            infer_script_path,
            "--cuda_idx", "0",
            "--stage1_model", "m-a-p/YuE-s1-7B-anneal-en-cot",
            "--stage2_model", "m-a-p/YuE-s2-1B-general",
            "--genre_txt", genre_file_path,
            "--lyrics_txt", lyrics_file_path,
            "--run_n_segments", "2",        # Full segments for better quality
            "--stage2_batch_size", "4",     # Higher batch size for speed
            "--output_dir", output_dir,
            "--max_new_tokens", "3000",     # Full token count
            "--profile", "1",               # Highest performance profile
            "--verbose", "3",
            "--rescale",                    # Enable audio rescaling to proper volume
            "--prompt_start_time", "0",
            "--prompt_end_time", "30",      # Full 30-second clips
        ]
        
        # Use flash attention if available, otherwise fallback
        if not flash_attn_available:
            cmd.append("--sdpa")
        
        # More detailed progress updates
        progress(0.1, desc="🚀 Initializing models...")
        progress(0.15, desc="📝 Processing lyrics...")
        progress(0.2, desc="🎵 Starting Stage 1 (7B model generation)...")
        
        # Extract parameters from cmd for logging
        run_n_segments = cmd[cmd.index("--run_n_segments") + 1] if "--run_n_segments" in cmd else "2"
        max_new_tokens = cmd[cmd.index("--max_new_tokens") + 1] if "--max_new_tokens" in cmd else "3000"
        
        print("🎵 Starting high-quality music generation...")
        print(f"📊 Generation settings: {run_n_segments} segments, {max_new_tokens} tokens, 30s audio")
        print(f"⏱️ Estimated time: 8-9 minutes for high-quality generation")
        print(f"Working directory: {os.getcwd()}")
        print(f"Command: {' '.join(cmd)}")
        
        # Change to YuEGP/inference directory for execution
        original_cwd = os.getcwd()
        inference_dir = os.path.join(os.getcwd(), "YuEGP", "inference")
        
        try:
            os.chdir(inference_dir)
            print(f"Changed to inference directory: {inference_dir}")
            cmd[1] = "infer.py"
            
            progress(0.25, desc="🔥 Stage 1: Running 7B parameter model...")
            
            # Start the subprocess
            import threading
            import time
            
            def parse_output_and_update_progress(process):
                """Parse subprocess output in real-time and update progress accordingly"""
                stage1_messages = [
                    "🧠 Stage 1: Generating musical concepts...",
                    "🎼 Stage 1: Creating melody patterns...",
                    "🎹 Stage 1: Composing harmony structure..."
                ]
                stage2_messages = [
                    "⚡ Starting Stage 2: Refining with 1B model...",
                    "🎵 Stage 2: Adding musical details...",
                    "🎶 Stage 2: Finalizing composition..."
                ]
                
                stage1_progress = [0.3, 0.45, 0.6]
                stage2_progress = [0.7, 0.8, 0.85]
                
                current_stage = 1
                stage1_step = 0
                stage2_step = 0
                
                output_lines = []
                
                try:
                    while True:
                        line = process.stdout.readline()
                        if not line:
                            break
                            
                        line = line.strip()
                        output_lines.append(line)
                        print(line)  # Still print for debugging
                        
                        # Check for stage transitions based on actual output
                        if "Stage 2 inference..." in line:
                            current_stage = 2
                            stage2_step = 0
                            progress(0.7, desc=stage2_messages[0])
                            print(f"⏳ {stage2_messages[0]}")
                        
                        elif "Stage 2 DONE" in line:
                            progress(0.9, desc="🔊 Decoding to audio format...")
                            print("⏳ 🔊 Decoding to audio format...")
                        
                        # Update Stage 1 progress periodically
                        elif current_stage == 1 and stage1_step < len(stage1_messages):
                            # Update Stage 1 progress every 15 seconds or on specific markers
                            if stage1_step < len(stage1_progress):
                                progress(stage1_progress[stage1_step], desc=stage1_messages[stage1_step])
                                print(f"⏳ {stage1_messages[stage1_step]}")
                                stage1_step += 1
                        
                        # Update Stage 2 progress periodically
                        elif current_stage == 2 and stage2_step < len(stage2_messages) - 1:
                            stage2_step += 1
                            if stage2_step < len(stage2_progress):
                                progress(stage2_progress[stage2_step], desc=stage2_messages[stage2_step])
                                print(f"⏳ {stage2_messages[stage2_step]}")
                
                except Exception as e:
                    print(f"Progress parsing error: {e}")
                
                return '\n'.join(output_lines)
            
            print(f"🚀 Executing command: {' '.join(cmd)}")
            
            # Use Popen for real-time output processing
            process = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, 
                                     text=True, bufsize=1, universal_newlines=True)
            
            # Parse output in real-time
            stdout_output = parse_output_and_update_progress(process)
            
            # Wait for process to complete and get return code
            return_code = process.wait()
            
            # Create result object similar to subprocess.run
            class Result:
                def __init__(self, returncode, stdout, stderr=""):
                    self.returncode = returncode
                    self.stdout = stdout
                    self.stderr = stderr
            
            result = Result(return_code, stdout_output)
            
            # Print stdout and stderr for debugging
            if result.stdout:
                print(f"✅ Command output:\n{result.stdout}")
            if result.stderr:
                print(f"⚠️ Command stderr:\n{result.stderr}")
            print(f"📊 Return code: {result.returncode}")
            
        finally:
            os.chdir(original_cwd)
        
        progress(0.95, desc="🎉 Processing completed, finalizing output...")
        
        # Clean up input files
        os.unlink(genre_file_path)
        os.unlink(lyrics_file_path)
        
        if result.returncode == 0:
            # Find generated audio file - prioritize mixed audio from vocoder/mix directory
            import glob

            final_files = glob.glob(os.path.join(output_dir, "*_mixed.mp3"))
            if final_files:
                progress(1.0, desc="Finish music generation")
                print(f"✅ Found audio file at root: {final_files[0]}")
                return final_files[0]

            # First look for the final mixed audio in vocoder/mix
            mixed_files = glob.glob(os.path.join(output_dir, "vocoder/mix/*_mixed.mp3"))
            if mixed_files:
                progress(1.0, desc="Music generation complete!")
                print(f"✅ Found mixed audio file: {mixed_files[0]}")
                return mixed_files[0]

            # Fallback to any MP3 file
            audio_files = glob.glob(os.path.join(output_dir, "**/*.mp3"), recursive=True)
            if audio_files:
                progress(1.0, desc="Music generation complete!")
                print(f"✅ Found audio file: {audio_files[0]}")
                return audio_files[0]  # Return path to generated audio
            else:
                print(f"❌ No audio files found in {output_dir}")
                print(f"Directory contents: {os.listdir(output_dir) if os.path.exists(output_dir) else 'Directory not found'}")
                return "Music generation completed but no audio file found."
        else:
            error_msg = f"Return code: {result.returncode}\n"
            if result.stderr:
                error_msg += f"Error: {result.stderr[-1000:]}\n"
            if result.stdout:
                error_msg += f"Output: {result.stdout[-1000:]}"
            return f"Music generation failed:\n{error_msg}"
            
    except subprocess.TimeoutExpired:
        return "Music generation timed out after 20 minutes. Please try again."
    except Exception as e:
        return f"Error during music generation: {str(e)}"

def respond(message, state):
    """Enhanced response function for conversational lyrics generation"""
    try:
        # Add user message to conversation
        state.conversation.append({"role": "user", "content": message})
        
        # Use conversational generation logic (same as voice input)
        response = generate_chat_completion(groq_client, state.conversation, state.genre, state.mood, state.theme)
        
        # Add assistant response
        state.conversation.append({"role": "assistant", "content": response})
        
        # Update lyrics with improved format recognition - extract only segments
        if any(marker in response.lower() for marker in ["[verse", "[chorus", "[bridge", "**verse", "**chorus", "sectiontype.verse", "verse:"]):
            state.lyrics = extract_lyrics_segments_only(response)
        
        # Format conversation for display
        return "", [{"role": msg["role"], "content": msg["content"]} for msg in state.conversation], state
        
    except Exception as e:
        error_response = f"Sorry, I encountered an error: {str(e)}"
        state.conversation.append({"role": "assistant", "content": error_response})
        return "", [{"role": msg["role"], "content": msg["content"]} for msg in state.conversation], state

def build_interface():
    """Build the Gradio interface optimized for Spaces with high performance"""
    
    with gr.Blocks(
        title="MiloMusic - AI Music Generation", 
        theme=gr.themes.Soft(),
        css="""
        .container { max-width: 1400px; margin: auto; }
        .performance-notice { background-color: #d4edda; padding: 15px; border-radius: 5px; margin: 10px 0; }
        .generation-status { background-color: #f8f9fa; padding: 10px; border-radius: 5px; }
        """
    ) as demo:
        
        # Header
        gr.Markdown("""
        # 🎵 MiloMusic - AI Music Generation
        ### Professional AI-powered music creation from natural language
        """)
        
        # Performance notice for Spaces
        gr.Markdown("""
        <div class="performance-notice">
        🚀 <strong>High-Performance Mode:</strong> Running on Spaces GPU with optimized settings for best quality.
        Generation time: ~8-9 minutes for professional-grade music with vocals and instruments.
        </div>
        """)
        
        state = gr.State(AppState())
        
        with gr.Row():
            with gr.Column(scale=2):
                # Input controls
                with gr.Group():
                    gr.Markdown("### 🎛️ Music Settings")
                    with gr.Row():
                        genre = gr.Dropdown(
                            choices=["pop", "rock", "jazz", "classical", "electronic", "folk", "r&b", "country", "hip-hop"],
                            value="pop", label="Genre"
                        )
                        mood = gr.Dropdown(
                            choices=["upbeat", "melancholic", "energetic", "calm", "romantic", "dark", "mysterious", "joyful"],
                            value="upbeat", label="Mood"
                        )
                        theme = gr.Dropdown(
                            choices=["love", "friendship", "adventure", "nostalgia", "freedom", "hope", "dreams", "nature"],
                            value="love", label="Theme"
                        )
                
                # Voice Input
                with gr.Group():
                    gr.Markdown("### 🎤 Voice Input")
                    input_audio = gr.Audio(
                        label="Speak Your Musical Ideas",
                        sources=["microphone"],
                        type="numpy",
                        streaming=False,
                        waveform_options=gr.WaveformOptions(waveform_color="#B83A4B"),
                    )
                
                # Chat interface
                with gr.Group():
                    gr.Markdown("### 💬 Lyrics Creation Chat")
                    chatbot = gr.Chatbot(height=400, label="AI Lyrics Assistant", show_copy_button=True, type="messages")
                    
                    with gr.Row():
                        text_input = gr.Textbox(
                            placeholder="Or type your song idea here...", 
                            show_label=False, 
                            scale=4,
                            lines=2
                        )
                        send_btn = gr.Button("Send", scale=1, variant="primary")
            
            with gr.Column(scale=1):
                # Output controls
                with gr.Group():
                    gr.Markdown("### 🎵 Music Generation")
                    lyrics_display = gr.Textbox(
                        label="Current Lyrics", 
                        lines=12, 
                        interactive=True,
                        placeholder="Your generated lyrics will appear here..."
                    )
                    
                    generate_btn = gr.Button("🎼 Generate High-Quality Music", variant="primary", size="lg")
                    
                    with gr.Column():
                        music_output = gr.Audio(label="Generated Music", type="filepath", show_download_button=True)
                        
                        gr.Markdown("""
                        <div class="generation-status">
                        <strong>Generation Features:</strong><br>
                        • Full 30-second clips<br>
                        • Professional vocals<br>
                        • Rich instrumentation<br>
                        • High-fidelity audio
                        </div>
                        """)
                
                # Controls
                with gr.Group():
                    gr.Markdown("### 🔧 Controls")
                    new_song_btn = gr.Button("🆕 Start New Song")
                    clear_btn = gr.Button("🧹 Clear Chat")
        
        # Event handlers
        def update_state_settings(genre_val, mood_val, theme_val, state):
            state.genre = genre_val
            state.mood = mood_val  
            state.theme = theme_val
            return state
        
        # Update state when settings change
        for component in [genre, mood, theme]:
            component.change(
                fn=update_state_settings,
                inputs=[genre, mood, theme, state],
                outputs=[state]
            )
        
        # Voice recording functionality (from app.py)
        stream = input_audio.start_recording(
            process_audio,
            [input_audio, state],
            [input_audio, state],
        )

        respond_audio = input_audio.stop_recording(
            response_audio, [state, input_audio, genre, mood, theme], [state, chatbot, lyrics_display]
        )

        restart = respond_audio.then(start_recording_user, [state], [input_audio]).then(
            lambda state: state, state, state, js=js_reset
        )
        
        # Text chat functionality with lyrics update
        def respond_with_lyrics_update(message, state):
            text_output, chat_output, updated_state = respond(message, state)
            return text_output, chat_output, updated_state, updated_state.lyrics
        
        send_btn.click(
            fn=respond_with_lyrics_update,
            inputs=[text_input, state],
            outputs=[text_input, chatbot, state, lyrics_display],
            queue=True
        )
        
        text_input.submit(
            fn=respond_with_lyrics_update,
            inputs=[text_input, state],
            outputs=[text_input, chatbot, state, lyrics_display],
            queue=True
        )
        
        # Music generation with progress
        generate_btn.click(
            fn=generate_music_spaces,
            inputs=[lyrics_display, genre, mood],
            outputs=[music_output],
            queue=True,
            show_progress=True
        )
        
        # Control buttons
        new_song_btn.click(
            fn=lambda: (AppState(), [], "", None, gr.Audio(recording=False)),
            outputs=[state, chatbot, lyrics_display, music_output, input_audio],
            cancels=[respond_audio, restart]
        )
        
        clear_btn.click(
            fn=lambda: [],
            outputs=[chatbot]
        )
        
        # Auto-update lyrics display when state changes
        state.change(
            fn=lambda s: s.lyrics,
            inputs=[state],
            outputs=[lyrics_display]
        )
        
        # Instructions
        gr.Markdown("""
        ### 📖 How to create your music:
        1. **Set your preferences**: Choose genre, mood, and theme
        2. **Voice or chat**: Either speak your ideas or type them in the chat
        3. **Refine the lyrics**: Ask for changes, different verses, or style adjustments  
        4. **Generate music**: Click the generate button for professional-quality output
        5. **Download & enjoy**: Your high-fidelity music with vocals and instruments
        
        **Tips**: Be specific about your vision - mention instruments, vocal style, or song structure!
        """)
        
        # Footer
        gr.Markdown("""
        ---
        <center>
        Made with ❤️ by the MiloMusic Team | Powered by YuE (乐) Model | 🤗 Hugging Face Spaces
        </center>
        """)
    
    return demo

# Audio transcription functions (from app.py)
def process_whisper_response(completion):
    """
    Process Whisper transcription response and filter out silence.
    """
    if completion.segments and len(completion.segments) > 0:
        no_speech_prob = completion.segments[0].get('no_speech_prob', 0)
        print("No speech prob:", no_speech_prob)

        if no_speech_prob > 0.7:
            print("No speech detected")
            return None

        return completion.text.strip()

    return None

def transcribe_audio(client, file_name):
    """
    Transcribe an audio file using the Whisper model via the Groq API.
    """
    if file_name is None:
        return None

    try:
        with open(file_name, "rb") as audio_file:
            with open("audio.wav", "wb") as f:
                f.write(audio_file.read())

            response = client.audio.transcriptions.create(
                model="whisper-large-v3-turbo",
                file=("audio.wav", audio_file),
                response_format="text",
                language="en",
            )
            
            # Process the response to filter out silence
            # For text response format, we need to check if response is meaningful
            if response and len(response.strip()) > 0:
                return response.strip()
            else:
                return None
                
    except Exception as e:
        print(f"Transcription error: {e}")
        return f"Error in audio transcription: {str(e)}"

def start_recording_user(state: AppState):
    """
    Reset the audio recording component for a new user input.
    """
    return None

def process_audio(audio: tuple, state: AppState):
    """
    Process recorded audio in real-time during recording.
    """
    return audio, state

@spaces.GPU(duration=40, progress=gr.Progress(track_tqdm=True))
def response_audio(state: AppState, audio: tuple, genre_value, mood_value, theme_value):
    """
    Process recorded audio and generate a response based on transcription.
    """
    if not audio:
        return state, []

    # Update state with current dropdown values
    state.genre, state.mood, state.theme = genre_value, mood_value, theme_value

    temp_dir = tempfile.gettempdir()
    file_name = os.path.join(temp_dir, f"{xxhash.xxh32(bytes(audio[1])).hexdigest()}.wav")

    sf.write(file_name, audio[1], audio[0], format="wav")

    api_key = os.environ.get("GROQ_API_KEY")
    if not api_key:
        raise ValueError("Please set the GROQ_API_KEY environment variable.")
    client = groq.Client(api_key=api_key)

    # Transcribe the audio file
    transcription = transcribe_audio(client, file_name)
    if transcription:
        if isinstance(transcription, str) and transcription.startswith("Error"):
            transcription = "Error in audio transcription."

        state.conversation.append({"role": "user", "content": transcription})

        assistant_message = generate_chat_completion(client, state.conversation, state.genre, state.mood, state.theme)

        state.conversation.append({"role": "assistant", "content": assistant_message})

        # Update lyrics using same logic as text input for consistency - extract only segments
        if any(marker in assistant_message.lower() for marker in ["[verse", "[chorus", "[bridge", "**verse", "**chorus", "sectiontype.verse", "verse:"]):
            state.lyrics = extract_lyrics_segments_only(assistant_message)

        os.remove(file_name)

    # Format conversation for display in messages format
    conversation_display = []
    for msg in state.conversation:
        conversation_display.append({"role": msg["role"], "content": msg["content"]})
    
    return state, conversation_display, state.lyrics

def extract_lyrics_segments_only(content):
    """
    Extract only the lyrics segments (VERSE, CHORUS, etc.) from AI response,
    removing any AI commentary or explanation text.
    """
    import re

    if not content:
        return ""

    lines = content.split('\n')
    lyrics_lines = []
    in_lyrics_section = False

    for line in lines:
        line = line.strip()

        # Check if this line is a section header (VERSE, CHORUS, etc.)
        if re.match(r'^\*\*(VERSE|CHORUS|BRIDGE|OUTRO).*\*\*$', line) or re.match(r'^\[(VERSE|CHORUS|BRIDGE|OUTRO).*\]$', line):
            in_lyrics_section = True
            lyrics_lines.append(line)
            continue

        # If we're in a lyrics section
        if in_lyrics_section:
            # Stop if we hit AI commentary
            if line and any(phrase in line.lower() for phrase in [
                'how do you like', 'would you like', 'let me know',
                'what do you think', 'any changes', 'take a look',
                'here are the lyrics', 'i\'ve created', 'feel free to'
            ]):
                break

            # Add lyrics line (including empty lines for formatting)
            lyrics_lines.append(line)

    return '\n'.join(lyrics_lines).strip()

def extract_lyrics_from_conversation(conversation):
    """
    Extract lyrics from conversation history with cross-platform compatibility.
    """
    lyrics = ""
    for message in reversed(conversation):
        if message["role"] == "assistant":
            content_lower = message["content"].lower()
            # 先尝试严格匹配（保持原逻辑）
            if "verse" in content_lower and "chorus" in content_lower:
                lyrics = extract_lyrics_segments_only(message["content"])
                break
            # 如果没找到，再用宽泛匹配（兼容性备选）
            elif any(marker in content_lower for marker in ["[verse", "[chorus", "**verse", "**chorus"]):
                lyrics = extract_lyrics_segments_only(message["content"])
                break
    return lyrics

def generate_chat_completion(client, history, genre, mood, theme):
    """
    Generate an AI assistant response based on conversation history and song parameters.
    """
    messages = []
    system_prompt = f"""You are a creative AI music generator assistant. Help users create song lyrics in the {genre} genre with a {mood} mood about {theme}.
When generating lyrics, create a chorus and at least one verse. Format lyrics clearly with VERSE and CHORUS labels.
Ask if they like the lyrics or want changes. Be conversational, friendly, and creative.
Keep the lyrics appropriate for the selected genre, mood, and theme unless the user specifically requests changes."""

    messages.append({
        "role": "system",
        "content": system_prompt,
    })

    for message in history:
        messages.append(message)

    try:
        completion = client.chat.completions.create(
            model="meta-llama/llama-4-scout-17b-16e-instruct",
            messages=messages,
        )
        return completion.choices[0].message.content
    except Exception as e:
        return f"Error in generating chat completion: {str(e)}"

# JavaScript for frontend enhancements
js_reset = """
() => {
  var record = document.querySelector('.record-button');
  if (record) {
    record.textContent = "Just Start Talking!"
    record.style = "width: fit-content; padding-right: 0.5vw;"
  }
}
"""

# Build the interface
demo = build_interface()

if __name__ == "__main__":
    """
    Spaces entry point - optimized for high-performance deployment
    """
    print("🚀 Starting MiloMusic High-Performance Mode on Hugging Face Spaces...")
    print(f"📁 Working directory: {os.getcwd()}")
    print(f"📂 Directory contents: {os.listdir('.')}")
    
    # Validate file structure
    if not validate_file_structure():
        print("❌ Required files missing - please check your upload")
        sys.exit(1)
    
    # Validate environment
    if not validate_api_keys():
        print("⚠️  Some API keys missing - functionality may be limited")
    
    # Launch with optimized settings for Spaces
    demo.queue(
        default_concurrency_limit=5,  # Allow more concurrent users
        max_size=20
    ).launch(
        server_name="0.0.0.0",
        server_port=7860,
        share=False,  # Spaces handles sharing
        show_error=True,
        quiet=False,
        favicon_path=None,
        ssl_verify=False
    )