app.py

"""
BELLO - Entornos Virtuales Afectivos (FULLY FIXED)
Working image generation + Enhanced 360° viewer with play/pause, continuous/random, and chunk dropdown
"""

import os
import math
import struct
import tempfile
import json
import base64
import warnings
import subprocess
import sys

# Suppress TensorFlow warnings
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
warnings.filterwarnings('ignore', category=UserWarning)
warnings.filterwarnings('ignore', category=DeprecationWarning)

try:
    from pydub import AudioSegment
except ImportError:
    print("Installing pydub...")
    subprocess.check_call([sys.executable, "-m", "pip", "install", "pydub"])
    from pydub import AudioSegment

import numpy as np
import gradio as gr
from PIL import Image, ImageDraw, ImageFont
import cv2

# Check for ffmpeg
try:
    subprocess.run(["ffmpeg", "-version"], capture_output=True, check=True)
except FileNotFoundError:
    print("⚠️  FFmpeg not found. Installing...")
    subprocess.check_call([sys.executable, "-m", "pip", "install", "ffmpeg-python"])

try:
    import torch
except ImportError:
    print("Installing torch...")
    subprocess.check_call([sys.executable, "-m", "pip", "install", "torch"])
    import torch

try:
    import scipy.io.wavfile
except ImportError:
    print("Installing scipy...")
    subprocess.check_call([sys.executable, "-m", "pip", "install", "scipy"])
    import scipy.io.wavfile

try:
    import requests
except ImportError:
    print("Installing requests...")
    subprocess.check_call([sys.executable, "-m", "pip", "install", "requests"])
    import requests

try:
    from textblob import TextBlob
except ImportError:
    print("Installing textblob...")
    subprocess.check_call([sys.executable, "-m", "pip", "install", "textblob"])
    from textblob import TextBlob

try:
    import librosa
except ImportError:
    print("Installing librosa...")
    subprocess.check_call([sys.executable, "-m", "pip", "install", "librosa"])
    import librosa

# =========================
# Model Loading with Fallbacks
# =========================

model = None
model2 = None
processor = None
music_model = None
device = None

def load_emotion_model(model_path):
    """Load emotion model with fallback"""
    try:
        from tensorflow.keras.models import load_model
        if os.path.exists(model_path):
            model = load_model(model_path)
            print(f"✅ Emotion model loaded: {model_path}")
            return model
        else:
            print(f"⚠️  Emotion model not found: {model_path}")
            return None
    except Exception as e:
        print(f"⚠️  Error loading emotion model: {e}")
        return None

model_path = "mymodel_SER_LSTM_RAVDESS.h5"
model = load_emotion_model(model_path)

# Try to load Whisper
try:
    from faster_whisper import WhisperModel
    print("📥 Loading Whisper model...")
    model2 = WhisperModel("small", device="cpu", compute_type="int8")
    print("✅ Whisper model loaded")
except Exception as e:
    print(f"⚠️  Whisper model not available: {e}")
    model2 = None

# Try to load MusicGen
def load_musicgen_model():
    try:
        print("📥 Loading MusicGen model...")
        from transformers import AutoProcessor, MusicgenForConditionalGeneration
        try:
            device = "cuda" if torch.cuda.is_available() else "cpu"
            print(f"Using device: {device}")
            processor = AutoProcessor.from_pretrained("facebook/musicgen-small")
            music_model = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small")
            music_model.to(device)
            # Set to eval mode to avoid gradient tracking
            music_model.eval()
            print("✅ MusicGen model loaded")
            return processor, music_model, device
        except Exception as load_error:
            print(f"⚠️  Error during MusicGen loading: {load_error}")
            return None, None, None
    except ImportError as e:
        print(f"⚠️  Transformers not installed: {e}")
        return None, None, None
    except Exception as e:
        print(f"⚠️  MusicGen model not available: {e}")
        return None, None, None

processor, music_model, device = load_musicgen_model()

# =========================
# Audio Processing
# =========================

def extract_mfcc(wav_filepath):
    """Extract MFCC features from audio"""
    try:
        y, sr = librosa.load(wav_filepath, sr=22050)
        mfcc = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=40)
        mfcc = np.mean(mfcc.T, axis=0)
        return mfcc
    except Exception as e:
        print(f"⚠️  MFCC extraction failed: {e}")
        return None

def transcribe(audio_path):
    """Transcribe audio using Whisper"""
    try:
        if model2 is None:
            return "[Transcripción no disponible]"
        
        segments, info = model2.transcribe(audio_path, language="es")
        text = "".join([segment.text for segment in segments])
        return text if text else "[Sin habla detectada]"
    except Exception as e:
        print(f"⚠️  Transcription error: {e}")
        return "[Error en transcripción]"

def chunk_audio_with_overlap(audio_path, chunk_duration=10, overlap_percent=20):
    """Split audio into chunks with overlap"""
    try:
        audio = AudioSegment.from_file(audio_path)
        duration_ms = len(audio)
        chunk_ms = chunk_duration * 1000
        overlap_ms = int(chunk_ms * (overlap_percent / 100.0))
        step_ms = chunk_ms - overlap_ms

        if chunk_duration <= 0:
            raise ValueError("Chunk duration must be positive")

        if chunk_duration > duration_ms / 1000:
            return [{"path": audio_path, "start_ms": 0, "end_ms": duration_ms, "original_index": 0, "overlap_ms": 0}], 1

        chunk_files = []
        num_chunks = math.ceil((duration_ms - overlap_ms) / step_ms) if step_ms > 0 else 1

        for i in range(num_chunks):
            start_ms = i * step_ms
            end_ms = min(start_ms + chunk_ms, duration_ms)
            
            if start_ms >= duration_ms:
                break
                
            chunk = audio[start_ms:end_ms]

            with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp_file:
                chunk.export(tmp_file.name, format="wav")
                chunk_files.append({
                    "path": tmp_file.name,
                    "start_ms": start_ms,
                    "end_ms": end_ms,
                    "original_index": i,
                    "overlap_ms": overlap_ms if i > 0 else 0
                })

        return chunk_files, len(chunk_files)

    except Exception as e:
        print(f"⚠️  Audio chunking error: {e}")
        try:
            audio_len = len(AudioSegment.from_file(audio_path))
            return [{"path": audio_path, "start_ms": 0, "end_ms": audio_len, "original_index": 0, "overlap_ms": 0}], 1
        except:
            return [], 0

# =========================
# Emotion & Sentiment Analysis
# =========================

emotions = {
    0: "neutral", 1: "calm", 2: "happy", 3: "sad",
    4: "angry", 5: "fearful", 6: "disgust", 7: "surprised",
}

def predict_emotion_from_audio(wav_filepath):
    """Predict emotion from audio"""
    try:
        if model is None:
            return "neutral"
        test_point = extract_mfcc(wav_filepath)
        if test_point is not None:
            test_point = np.reshape(test_point, newshape=(1, 40, 1))
            predictions = model.predict(test_point, verbose=0)
            predicted_class = np.argmax(predictions[0])
            return emotions.get(predicted_class, "neutral")
        return "neutral"
    except Exception as e:
        print(f"⚠️  Emotion prediction error: {e}")
        return "neutral"

def analyze_sentiment(text):
    """Analyze sentiment from text"""
    try:
        if not text or text.strip() == "" or text.startswith("["):
            return "neutral", 0.0
        analysis = TextBlob(text)
        polarity = analysis.sentiment.polarity
        sentiment = "positive" if polarity > 0.1 else "negative" if polarity < -0.1 else "neutral"
        return sentiment, polarity
    except Exception as e:
        print(f"⚠️  Sentiment analysis error: {e}")
        return "neutral", 0.0

# =========================
# Image Generation (Fixed - Placeholder based on Sentiment)
# =========================

def generate_image(sentiment_prediction, transcribed_text, chunk_idx, total_chunks):
    """
    Generate a 360° panorama placeholder image based on sentiment.
    Creates colored equirectangular image with text overlay.
    """
    try:
        print(f"🎨 Generating image for segment {chunk_idx + 1}, sentiment: {sentiment_prediction}")
        
        # Color mapping based on sentiment
        color_map = {
            "positive": (100, 200, 100),    # Green (RGB)
            "negative": (200, 100, 100),    # Red (RGB)
            "neutral": (100, 150, 200),     # Blue (RGB)
        }
        
        # Get color with fallback
        color = color_map.get(sentiment_prediction.lower(), (100, 150, 200))
        print(f"   Using color: RGB{color}")
        
        # Create equirectangular image (2:1 aspect ratio for 360° panorama)
        print(f"   Creating image: 1024x512")
        image = Image.new("RGB", (1024, 512), color=color)
        
        if image is None:
            raise ValueError("Image creation returned None")
        
        print(f"   Image created successfully: {image.size}")
        
        # Add text overlay
        try:
            from PIL import ImageDraw, ImageFont
            draw = ImageDraw.Draw(image)
            
            # Try to use a nice font, fallback to default
            try:
                font = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf", 20)
            except:
                font = ImageFont.load_default()
            
            # Add sentiment text
            text = f"Segment {chunk_idx + 1} - {sentiment_prediction.upper()}"
            print(f"   Adding text: {text}")
            
            # Calculate center position
            bbox = draw.textbbox((0, 0), text, font=font)
            text_width = bbox[2] - bbox[0]
            text_x = (1024 - text_width) // 2
            text_y = 240
            
            # Draw text with white color
            draw.text((text_x, text_y), text, fill=(255, 255, 255), font=font)
            print(f"   Text added at position ({text_x}, {text_y})")
            
            # Add transcription preview (first 60 chars)
            if transcribed_text and not transcribed_text.startswith("["):
                preview = transcribed_text[:60] + "..." if len(transcribed_text) > 60 else transcribed_text
                preview_bbox = draw.textbbox((0, 0), preview, font=font)
                preview_width = preview_bbox[2] - preview_bbox[0]
                preview_x = (1024 - preview_width) // 2
                preview_y = 280
                draw.text((preview_x, preview_y), preview, fill=(200, 200, 200), font=font)
                print(f"   Preview added: {preview[:40]}...")
        except Exception as text_error:
            print(f"⚠️  Text overlay error (non-critical): {text_error}")
            # Continue even if text fails
        
        print(f"✅ Image generated successfully for segment {chunk_idx + 1}")
        return image
        
    except Exception as e:
        print(f"❌ Image generation error: {e}")
        import traceback
        traceback.print_exc()
        # Return fallback image
        fallback = Image.new("RGB", (1024, 512), color=(100, 150, 200))
        print(f"   Using fallback image")
        return fallback

def create_xmp_block(width, height):
    """Create XMP metadata block for 360° panorama"""
    xmp = (
        f'<?xpacket begin="" id="W5M0MpCehiHzreSzNTczkc9d"?>\n'
        f'<x:xmpmeta xmlns:x="adobe:ns:meta/" x:xmptk="ExifTool">\n'
        f'<rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#">\n'
        f'<rdf:Description rdf:about=""\n'
        f'xmlns:GPano="http://ns.google.com/photos/1.0/panorama/"\n'
        f'GPano:ProjectionType="equirectangular"\n'
        f'GPano:UsePanoramaViewer="True"\n'
        f'GPano:FullPanoWidthPixels="{width}"\n'
        f'GPano:FullPanoHeightPixels="{height}"\n'
        f'GPano:CroppedAreaImageWidthPixels="{width}"\n'
        f'GPano:CroppedAreaImageHeightPixels="{height}"\n'
        f'GPano:CroppedAreaLeftPixels="0"\n'
        f'GPano:CroppedAreaTopPixels="0"/>\n'
        f'</rdf:RDF>\n'
        f'</x:xmpmeta>\n'
        f'<?xpacket end="w"?>'
    )
    return xmp

def write_xmp_to_jpg(input_path, output_path, width, height):
    """Inject XMP 360° metadata into JPEG"""
    try:
        with open(input_path, "rb") as f:
            data = f.read()

        if data[0:2] != b"\xFF\xD8":
            raise ValueError("Not a valid JPEG file")

        xmp_data = create_xmp_block(width, height)
        app1_marker = b"\xFF\xE1"
        xmp_header = b"http://ns.adobe.com/xap/1.0/\x00"
        xmp_bytes = xmp_data.encode("utf-8")
        length = len(xmp_header) + len(xmp_bytes) + 2
        length_bytes = struct.pack(">H", length)

        output = bytearray()
        output.extend(data[0:2])  # SOI
        output.extend(app1_marker)
        output.extend(length_bytes)
        output.extend(xmp_header)
        output.extend(xmp_bytes)
        output.extend(data[2:])

        with open(output_path, "wb") as f:
            f.write(output)
    except Exception as e:
        print(f"⚠️  XMP write error: {e}")

def add_360_metadata(img):
    """Add 360° panorama metadata to image and save"""
    try:
        print(f"   Saving image with metadata...")
        
        if img is None:
            raise ValueError("Input image is None")
        
        # Save to temporary file first
        with tempfile.NamedTemporaryFile(suffix=".jpg", delete=False) as tmp_file:
            tmp_path = tmp_file.name
            print(f"   Temp file: {tmp_path}")
            
            # Ensure image is in RGB mode
            if img.mode != 'RGB':
                print(f"   Converting from {img.mode} to RGB")
                img = img.convert('RGB')
            
            # Save as JPEG
            img.save(tmp_path, "JPEG", quality=95)
            print(f"   JPEG saved: {tmp_path}")
            
            # Inject XMP metadata
            try:
                write_xmp_to_jpg(tmp_path, tmp_path, img.width, img.height)
                print(f"   XMP metadata injected")
            except Exception as xmp_error:
                print(f"   ⚠️  XMP injection failed (non-critical): {xmp_error}")
                # Continue even if XMP fails
            
            print(f"✅ Image saved: {tmp_path}")
            return tmp_path
            
    except Exception as e:
        print(f"❌ Metadata error: {e}")
        import traceback
        traceback.print_exc()
        
        # Fallback: save without metadata
        try:
            with tempfile.NamedTemporaryFile(suffix=".jpg", delete=False) as tmp_file:
                tmp_path = tmp_file.name
                if img.mode != 'RGB':
                    img = img.convert('RGB')
                img.save(tmp_path, "JPEG", quality=95)
                print(f"⚠️  Saved without metadata: {tmp_path}")
                return tmp_path
        except Exception as fallback_error:
            print(f"❌ Fallback also failed: {fallback_error}")
            return None

# =========================
# Music Generation
# =========================

def generate_music(transcribed_text, emotion_prediction, chunk_idx, total_chunks):
    """Generate music using MusicGen (if available)"""
    try:
        if processor is None or music_model is None:
            print("⚠️  MusicGen not available, skipping music generation")
            return None
        
        emotion_prompts = {
            "calm": "calm relaxing ambient music",
            "happy": "uplifting energetic joyful music",
            "sad": "melancholic emotional sad music",
            "angry": "intense dramatic aggressive music",
            "fearful": "eerie suspenseful music",
            "disgust": "dark unsettling music",
            "surprised": "playful expressive music",
            "neutral": "ambient peaceful background music"
        }
        
        prompt = emotion_prompts.get(emotion_prediction.lower(), "ambient background music")
        
        try:
            with torch.no_grad():
                inputs = processor(text=[prompt], padding=True, return_tensors="pt").to(device)
                audio_values = music_model.generate(**inputs, max_new_tokens=256)
            
            sampling_rate = music_model.config.audio_encoder.sampling_rate
            audio_data = audio_values[0, 0].cpu().numpy()
            audio_data = audio_data / max(1e-9, np.max(np.abs(audio_data)))
            
            with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp_file:
                scipy.io.wavfile.write(tmp_file.name, rate=sampling_rate, data=audio_data)
                print(f"✅ Music generated for segment {chunk_idx + 1}")
                return tmp_file.name
        except RuntimeError as runtime_error:
            print(f"⚠️  Music generation runtime error: {runtime_error}")
            print("   Skipping music for this segment")
            return None
            
    except Exception as e:
        print(f"⚠️  Music generation error: {e}")
        return None

# =========================
# Chunk Processing
# =========================

def process_chunk(chunk_info, generate_audio=True):
    """Process a single audio chunk"""
    try:
        chunk_path = chunk_info["path"]
        chunk_idx = chunk_info["original_index"]
        
        print(f"\n📋 ====== Processing Chunk {chunk_idx + 1} ======")
        
        print(f"🎯 Step 1: Emotion prediction...")
        emotion_prediction = predict_emotion_from_audio(chunk_path)
        print(f"   ✓ Emotion: {emotion_prediction}")
        
        print(f"🎯 Step 2: Transcription...")
        transcribed_text = transcribe(chunk_path)
        print(f"   ✓ Text: {transcribed_text[:60]}..." if len(transcribed_text) > 60 else f"   ✓ Text: {transcribed_text}")
        
        print(f"🎯 Step 3: Sentiment analysis...")
        sentiment, polarity = analyze_sentiment(transcribed_text)
        print(f"   ✓ Sentiment: {sentiment} (polarity: {polarity:.2f})")
        
        print(f"🎯 Step 4: Image generation...")
        image = generate_image(sentiment, transcribed_text, chunk_idx, -1)
        if image is None:
            raise ValueError("Image generation returned None")
        print(f"   ✓ Image: {image.size} {image.mode}")
        
        print(f"🎯 Step 5: Adding 360° metadata...")
        image_360_path = add_360_metadata(image)
        print(f"   ✓ Path: {image_360_path}")
        
        music_path = None
        if generate_audio:
            print(f"🎯 Step 6: Music generation...")
            music_path = generate_music(transcribed_text, emotion_prediction, chunk_idx, -1)
            if music_path:
                print(f"   ✓ Music: {music_path}")
            else:
                print(f"   ⚠️  Music skipped")

        result = {
            "chunk_index": chunk_idx + 1,
            "emotion": emotion_prediction,
            "transcription": transcribed_text[:100] if transcribed_text else "N/A",
            "sentiment": sentiment,
            "image": image,
            "image_360": image_360_path,
            "music": music_path,
        }
        
        print(f"✅ Chunk {chunk_idx + 1} processed successfully")
        return result
        
    except Exception as e:
        print(f"\n❌ ERROR processing chunk {chunk_idx + 1}: {e}")
        import traceback
        traceback.print_exc()
        return {
            "chunk_index": chunk_idx + 1,
            "emotion": "error",
            "transcription": str(e),
            "sentiment": "error",
            "image": Image.new("RGB", (1024, 512), color=(100, 100, 100)),
            "image_360": None,
            "music": None,
        }

def get_predictions(audio_input, generate_audio=True, chunk_duration=10):
    """Process all chunks"""
    try:
        chunk_infos, total_chunks = chunk_audio_with_overlap(audio_input, chunk_duration, overlap_percent=20)
        results = []
        
        for i, chunk_info in enumerate(chunk_infos):
            print(f"⏳ Processing chunk {i+1}/{total_chunks}")
            result = process_chunk(chunk_info, generate_audio)
            results.append(result)
        
        # Cleanup temp files
        for chunk_info in chunk_infos:
            try:
                if chunk_info["path"] != audio_input:
                    os.unlink(chunk_info["path"])
            except:
                pass
        
        return results
    except Exception as e:
        print(f"⚠️  Prediction error: {e}")
        return []

# =========================
# 360 Viewer - ENHANCED
# =========================

def create_360_viewer_html(image_paths, audio_paths, output_path):
    """Create enhanced 360 viewer with play/pause, continuous/random, chunk dropdown"""
    image_data_list = []
    for img_path in image_paths:
        try:
            with open(img_path, "rb") as f:
                img_data = base64.b64encode(f.read()).decode("utf-8")
                image_data_list.append(f"data:image/jpeg;base64,{img_data}")
        except Exception as e:
            print(f"⚠️  Image encoding error: {e}")
            image_data_list.append(None)

    audio_data_list = []
    for audio_path in audio_paths:
        if audio_path:
            try:
                with open(audio_path, "rb") as f:
                    audio_data = base64.b64encode(f.read()).decode("utf-8")
                    audio_data_list.append(f"data:audio/wav;base64,{audio_data}")
            except Exception as e:
                print(f"⚠️  Audio encoding error: {e}")
                audio_data_list.append(None)
        else:
            audio_data_list.append(None)

    html_content = f"""
<!DOCTYPE html>
<html lang="es">
<head>
  <meta charset="UTF-8">
  <meta name="viewport" content="width=device-width, initial-scale=1.0">
  <title>EVA 360 - Visualizador Afectivo</title>
  <link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/pannellum@2.5.6/build/pannellum.css"/>
  <style>
    * {{ margin: 0; padding: 0; box-sizing: border-box; }}
    body {{
      overflow: hidden;
      font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif;
      background: #0a0e27;
      color: #e0e0e0;
    }}
    #panorama {{
      width: 100vw;
      height: calc(100vh - 120px);
    }}
    #controls {{
      position: fixed;
      bottom: 0;
      left: 0;
      right: 0;
      height: 120px;
      background: linear-gradient(to top, rgba(15, 25, 55, 0.98), rgba(15, 25, 55, 0.90));
      border-top: 2px solid rgba(100, 200, 255, 0.3);
      padding: 15px 20px;
      display: flex;
      gap: 20px;
      align-items: center;
      justify-content: flex-start;
      z-index: 1000;
      flex-wrap: wrap;
    }}
    .control-group {{
      display: flex;
      align-items: center;
      gap: 10px;
    }}
    .control-label {{
      color: #80b0ff;
      font-weight: 600;
      font-size: 12px;
      text-transform: uppercase;
      margin-right: 5px;
    }}
    select, input[type="range"] {{
      padding: 8px 12px;
      background: rgba(20, 40, 70, 0.9);
      border: 1px solid rgba(150, 200, 255, 0.3);
      color: #e0e0e0;
      border-radius: 6px;
      font-size: 13px;
      cursor: pointer;
    }}
    select:focus, input[type="range"]:focus {{
      outline: none;
      border-color: rgba(100, 180, 255, 0.6);
      background: rgba(30, 50, 80, 0.95);
    }}
    button {{
      padding: 10px 16px;
      background: rgba(80, 150, 200, 0.6);
      border: 1px solid rgba(150, 200, 255, 0.3);
      color: #e0e0e0;
      border-radius: 6px;
      cursor: pointer;
      transition: all 0.3s ease;
      font-weight: 600;
      font-size: 13px;
    }}
    button:hover {{
      background: rgba(100, 180, 255, 0.8);
      border-color: rgba(150, 200, 255, 0.6);
    }}
    button.active {{
      background: #2ecc71;
      border-color: #2ecc71;
      color: #fff;
    }}
    .toggle-group {{
      display: flex;
      gap: 5px;
    }}
    .toggle-group button {{
      padding: 8px 12px;
      font-size: 12px;
    }}
    #chunk-selector {{
      min-width: 150px;
    }}
  </style>
</head>
<body>
  <div id="panorama"></div>
  
  <div id="controls">
    <div class="control-group">
      <button id="playBtn" class="active">▶ Reproducir</button>
      <button id="pauseBtn">⏸ Pausa</button>
    </div>
    
    <div class="control-group">
      <div class="toggle-group">
        <button id="continuousBtn" class="active" title="Reproducir secuencialmente">🔄 Continuo</button>
        <button id="randomBtn" title="Reproducir en orden aleatorio">🔀 Aleatorio</button>
      </div>
    </div>
    
    <div class="control-group">
      <label class="control-label">Saltar a Segmento:</label>
      <select id="chunk-selector">
        {"".join([f'<option value="{i}">Segmento {i+1}</option>' for i in range(len(image_data_list))])}
      </select>
    </div>
    
    <div class="control-group" style="margin-left: auto;">
      <span id="status" style="font-size: 12px; color: #b0b0ff;">Parado</span>
    </div>
  </div>
  <script src="https://cdn.jsdelivr.net/npm/pannellum@2.5.6/build/pannellum.js"></script>
  <script>
    const images = {json.dumps(image_data_list)};
    const audioFiles = {json.dumps(audio_data_list)};
    
    let state = {{
      currentIndex: 0,
      mode: 'continuous',
      isPlaying: false,
      viewer: null,
      sequence: []
    }};
    
    function loadPanorama(index) {{
      if (state.viewer) {{
        try {{ state.viewer.destroy(); }} catch(e) {{}}
      }}
      state.currentIndex = index;
      document.getElementById('chunk-selector').value = index;
      
      state.viewer = pannellum.viewer('panorama', {{
        type: "equirectangular",
        panorama: images[index],
        autoLoad: true,
        autoRotate: -2,
        showZoomCtrl: true,
        showFullscreenCtrl: true,
        hfov: 100
      }});
    }}
    
    async function playSequence() {{
      state.isPlaying = true;
      document.getElementById('playBtn').classList.add('active');
      document.getElementById('pauseBtn').classList.remove('active');
      document.getElementById('status').textContent = 'Reproduciendo...';
      
      state.sequence = Array.from({{length: images.length}}, (_, i) => i);
      
      if (state.mode === 'random') {{
        // Fisher-Yates shuffle
        for (let i = state.sequence.length - 1; i > 0; i--) {{
          const j = Math.floor(Math.random() * (i + 1));
          [state.sequence[i], state.sequence[j]] = [state.sequence[j], state.sequence[i]];
        }}
      }}
      
      for (let idx of state.sequence) {{
        if (!state.isPlaying) {{
          document.getElementById('status').textContent = 'Pausado';
          break;
        }}
        
        loadPanorama(idx);
        
        if (audioFiles[idx]) {{
          try {{
            const audio = new Audio(audioFiles[idx]);
            document.getElementById('status').textContent = `Reproduciendo segmento ${{idx + 1}}...`;
            await new Promise(resolve => {{
              audio.onended = resolve;
              audio.play().catch(e => {{
                console.log('Play error:', e);
                setTimeout(resolve, 2000);
              }});
            }});
          }} catch (e) {{
            console.log('Audio error:', e);
            await new Promise(r => setTimeout(r, 2000));
          }}
        }} else {{
          document.getElementById('status').textContent = `Segmento ${{idx + 1}} (sin audio)`;
          await new Promise(r => setTimeout(r, 2000));
        }}
      }}
      
      if (state.isPlaying) {{
        state.isPlaying = false;
        document.getElementById('playBtn').classList.remove('active');
        document.getElementById('status').textContent = 'Completado';
      }}
    }}
    
    // Play button
    document.getElementById('playBtn').addEventListener('click', function() {{
      if (!state.isPlaying) {{
        playSequence();
      }}
    }});
    
    // Pause button
    document.getElementById('pauseBtn').addEventListener('click', function() {{
      state.isPlaying = false;
      document.getElementById('playBtn').classList.remove('active');
      document.getElementById('status').textContent = 'Pausado';
    }});
    
    // Mode toggles
    document.getElementById('continuousBtn').addEventListener('click', function() {{
      state.mode = 'continuous';
      this.classList.add('active');
      document.getElementById('randomBtn').classList.remove('active');
      document.getElementById('status').textContent = 'Modo: Continuo';
    }});
    
    document.getElementById('randomBtn').addEventListener('click', function() {{
      state.mode = 'random';
      this.classList.add('active');
      document.getElementById('continuousBtn').classList.remove('active');
      document.getElementById('status').textContent = 'Modo: Aleatorio';
    }});
    
    // Chunk selector dropdown
    document.getElementById('chunk-selector').addEventListener('change', function(e) {{
      const selectedIndex = parseInt(e.target.value);
      state.isPlaying = false;
      loadPanorama(selectedIndex);
      document.getElementById('status').textContent = `Segmento ${{selectedIndex + 1}}`;
    }});
    
    // Initialize
    loadPanorama(0);
    document.getElementById('status').textContent = 'Listo para reproducir';
  </script>
</body>
</html>
"""
    with open(output_path, "w") as f:
        f.write(html_content)
    return output_path

# =========================
# Gradio Interface
# =========================

def process_audio(audio_input, generate_audio, chunk_duration):
    """Main processing function"""
    if not audio_input:
        return "❌ Por favor carga un archivo de audio", [], None
    
    # Validate and fix chunk duration
    if chunk_duration is None:
        chunk_duration = 10
    else:
        try:
            chunk_duration = float(chunk_duration)
            if chunk_duration < 1:
                chunk_duration = 10
                print(f"⚠️  Chunk duration too small, using default: 10s")
            elif chunk_duration > 120:
                chunk_duration = 120
                print(f"⚠️  Chunk duration too large, using maximum: 120s")
        except (ValueError, TypeError):
            chunk_duration = 10
            print(f"⚠️  Invalid chunk duration, using default: 10s")
    
    print(f"\n🚀 Starting processing with {chunk_duration}s chunks...")
    
    results = get_predictions(audio_input, generate_audio, chunk_duration)
    
    if not results:
        return "❌ Error al procesar el audio", [], None
    
    # Build output markdown
    output_md = f"## ✅ Procesamiento Completado\n\n"
    output_md += f"**Total de segmentos:** {len(results)}\n\n"
    
    images_list = []
    
    for i, result in enumerate(results):
        output_md += f"### 📊 Segmento {i+1}\n"
        output_md += f"- **Emoción:** {result['emotion']}\n"
        output_md += f"- **Transcripción:** {result['transcription']}\n"
        output_md += f"- **Sentimiento:** {result['sentiment']}\n\n"
        
        if result['image']:
            images_list.append((result['image'], f"Seg {i+1}"))
    
    # Create viewer
    viewer_path = None
    image_paths = [r['image_360'] for r in results if r['image_360']]
    audio_paths = [r['music'] for r in results]
    
    if image_paths:
        try:
            with tempfile.NamedTemporaryFile(suffix=".html", delete=False, mode='w') as tmp:
                viewer_path = create_360_viewer_html(image_paths, audio_paths, tmp.name)
                print(f"✅ Viewer created: {viewer_path}")
        except Exception as e:
            print(f"⚠️  Error creating viewer: {e}")
            viewer_path = None
    
    return output_md, images_list, viewer_path

# Create interface
with gr.Blocks(title="BELLO Seamless Enhanced") as demo:
    gr.Markdown("# 🌀 BELLO Seamless - Entornos Virtuales Afectivos")
    gr.Markdown("""
    **BELLO Seamless** explora las emociones a través de la voz y crea experiencias inmersivas en 360°.
    
    ### ¿Cómo usar?
    1. 🎤 Carga un archivo de audio
    2. ⚙️ Ajusta la duración de segmentos (5-60 segundos)
    3. 🎵 Marca para generar música (opcional, más lento)
    4. 🚀 Presiona "Generar EVA"
    5. 📥 Descarga y abre el visualizador HTML en tu navegador
    6. ▶️ Usa los controles: Play/Pause, Continuo/Aleatorio, Dropdown de segmentos
    """)

    with gr.Row():
        with gr.Column(scale=2):
            audio = gr.Audio(label="🎤 Audio", type="filepath", sources=["upload", "microphone"])
        
        with gr.Column(scale=1):
            duration = gr.Number(label="⏱ Duración Segmento (s)", value=10, minimum=2, maximum=120, step=1)
            gen_music = gr.Checkbox(label="🎵 Generar Música", value=False)

    with gr.Row():
        process_btn = gr.Button("🚀 Generar EVA", variant="primary")
        clear_btn = gr.Button("🗑 Limpiar", variant="secondary")

    with gr.Tabs():
        with gr.TabItem("📊 Resultados"):
            output_text = gr.Markdown(value="*Resultados aparecerán aquí*")
            gallery = gr.Gallery(label="🖼 Imágenes", columns=2, rows=2)
        
        with gr.TabItem("🎬 Visualizador 360°"):
            gr.Markdown("### 📥 Descarga tu EVA aquí")
            gr.Markdown("Una vez que el procesamiento esté completo, haz clic en el botón de descarga para obtener el visualizador HTML interactivo.")
            viewer = gr.File(label="📥 Descargar Visualizador 360°", type="filepath")

    def clear():
        return None, "❌ Borrado", [], None

    process_btn.click(
        fn=process_audio,
        inputs=[audio, gen_music, duration],
        outputs=[output_text, gallery, viewer]
    )

    clear_btn.click(
        fn=clear,
        outputs=[audio, output_text, gallery, viewer]
    )

if __name__ == "__main__":
    print("\n" + "="*60)
    print("🌀 BELLO Seamless Enhanced - Iniciando...")
    print("="*60 + "\n")
    
    demo.launch(
        server_name="0.0.0.0",
        server_port=7860,
        share=False,
        show_error=True
    )