import os
import json
import traceback
import logging
import gradio as gr
import numpy as np
import librosa
import torch
import asyncio
import edge_tts
import re
import shutil
import time
from datetime import datetime
from fairseq import checkpoint_utils
from fairseq.data.dictionary import Dictionary
from lib.infer_pack.models import (
    SynthesizerTrnMs256NSFsid,
    SynthesizerTrnMs256NSFsid_nono,
    SynthesizerTrnMs768NSFsid,
    SynthesizerTrnMs768NSFsid_nono,
)
from vc_infer_pipeline import VC
from config import Config

# =============================
# LOAD ENVIRONMENT VARIABLES
# =============================
HF_TOKEN = os.getenv("HF_TOKEN")
if HF_TOKEN:
    print("🔑 Hugging Face token detected")
    os.environ["HUGGINGFACE_TOKEN"] = HF_TOKEN
else:
    print("⚠️ No HF_TOKEN found")

# =============================
# DOWNLOAD WEIGHTS OTOMATIS
# =============================
if not os.path.exists("weights"):
    print("=" * 50)
    print("🚀 BANGO DREAM MYGO VOICE CONVERSION")
    print("=" * 50)
    print("📥 Mendownload weights dan bahan model dari repo Plana-RCV/BanGDream-MyGO...")
    
    try:
        from huggingface_hub import snapshot_download
        
        repo_id = "Plana-Archive/Premium-Model"
        print(f"📥 Downloading from: {repo_id}")
        print("📁 Looking for: BanGDream-MyGO")
        
        downloaded_path = snapshot_download(
            repo_id=repo_id,
            allow_patterns=[
                "BanGDream-MyGO/weights/**",
                "BanGDream-MyGO/hubert_base.pt",
                "BanGDream-MyGO/rmvpe.pt"
            ],
            local_dir=".",
            local_dir_use_symlinks=False,
            token=HF_TOKEN if HF_TOKEN else None,
            max_workers=2
        )
        
        print("✅ Download completed")
        
        source_dir = "BanGDream-MyGO"
        if os.path.exists(source_dir):
            print(f"📂 Moving files from: {source_dir}")
            for item in os.listdir(source_dir):
                s = os.path.join(source_dir, item)
                d = os.path.join(".", item)
                if os.path.isdir(s):
                    if os.path.exists(d):
                        shutil.rmtree(d)
                    shutil.move(s, d)
                else:
                    shutil.move(s, d)
            shutil.rmtree(source_dir)
            print("✅ Files moved successfully")
            
            folder_info_path = os.path.join("weights", "folder_info.json")
            if not os.path.exists(folder_info_path):
                folder_info = {
                    "BanGDream-MyGO": {
                        "title": "BanG Dream! MyGO!!!!!",
                        "folder_path": "BanGDream-MyGO",
                        "description": "Official RVC Weights for BanG Dream! MyGO!!!!! characters",
                        "enable": True
                    }
                }
                with open(folder_info_path, "w", encoding="utf-8") as f:
                    json.dump(folder_info, f, indent=2, ensure_ascii=False)
                print(f"📄 Created folder_info.json")
        else:
            print("❌ Source directory not found after download!")
                
    except Exception as e:
        print(f"⚠️ Download failed: {str(e)}")
        traceback.print_exc()
        print("\n📝 Manual setup:")
        print("1. Create folder: weights/")
        print("2. Download from: https://huggingface.co/Plana-Archive/Plana-RCV/tree/main/BanGDream-MyGO")
        print("3. Put BanGDream-MyGO folder in weights/")

config = Config()
logging.getLogger("numba").setLevel(logging.WARNING)
logging.getLogger("fairseq").setLevel(logging.WARNING)

model_cache = {}
hubert_loaded = False
hubert_model = None

spaces = True 
if spaces:
    audio_mode = ["Upload audio", "TTS Audio"]
else:
    audio_mode = ["Input path", "Upload audio", "TTS Audio"]

f0method_mode = ["pm", "harvest"]
if os.path.isfile("rmvpe.pt"):
    f0method_mode.insert(2, "rmvpe")

def clean_title(title):
    title = re.sub(r'^BanG Dream[!]?\s*MyGO[!]*\s*-\s*', '', title, flags=re.IGNORECASE)
    return re.sub(r'\s*-\s*\d+\s*epochs', '', title, flags=re.IGNORECASE)

def _load_audio_input(vc_audio_mode, vc_input, vc_upload, tts_text, spaces_limit=20):
    temp_file = None
    try:
        if vc_audio_mode == "Input path" and vc_input:
            audio, sr = librosa.load(vc_input, sr=16000, mono=True)
            return audio.astype(np.float32), 16000, None
            
        elif vc_audio_mode == "Upload audio":
            if vc_upload is None:
                raise ValueError("Mohon upload file audio terlebih dahulu!")
            sampling_rate, audio = vc_upload
            
            if audio.dtype != np.float32:
                audio = audio.astype(np.float32) / np.iinfo(audio.dtype).max
            
            if len(audio.shape) > 1:
                audio = np.mean(audio, axis=0)
                
            if sampling_rate != 16000:
                audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=16000, res_type='kaiser_fast')
                
            return audio.astype(np.float32), 16000, None
            
        elif vc_audio_mode == "TTS Audio":
            if not tts_text or tts_text.strip() == "":
                raise ValueError("Mohon masukkan teks untuk TTS!")
            
            temp_file = "tts_temp.wav"
            async def tts_task():
                return await edge_tts.Communicate(tts_text, "ja-JP-NanamiNeural").save(temp_file)
            
            try:
                asyncio.run(asyncio.wait_for(tts_task(), timeout=10))
            except asyncio.TimeoutError:
                raise ValueError("TTS timeout! Silakan coba lagi.")
            
            audio, sr = librosa.load(temp_file, sr=16000, mono=True)
            return audio.astype(np.float32), 16000, temp_file
            
    except Exception as e:
        if temp_file and os.path.exists(temp_file):
            os.remove(temp_file)
        raise e
        
    raise ValueError("Invalid audio mode or missing input.")

def adjust_audio_speed(audio, speed):
    if speed == 1.0:
        return audio
    return librosa.effects.time_stretch(audio.astype(np.float32), rate=speed)

def preprocess_audio(audio):
    if np.max(np.abs(audio)) > 1.0:
        audio = audio / np.max(np.abs(audio)) * 0.9
    return audio.astype(np.float32)

def create_vc_fn(model_key, tgt_sr, net_g, vc, if_f0, version, file_index):
    def vc_fn(
        vc_audio_mode, vc_input, vc_upload, tts_text,
        f0_up_key, f0_method, index_rate, filter_radius, 
        resample_sr, rms_mix_rate, protect, speed,
    ):
        temp_audio_file = None
        try:
            if torch.cuda.is_available():
                torch.cuda.empty_cache()
                
            net_g.to(config.device)
            yield "Status: 🚀 Memproses audio...", None
            
            audio, sr, temp_audio_file = _load_audio_input(vc_audio_mode, vc_input, vc_upload, tts_text)
            audio = preprocess_audio(audio)
            audio_tensor = torch.FloatTensor(audio).to(config.device)
            times = [0, 0, 0]
            
            max_chunk_size = 16000 * 30
            if len(audio) > max_chunk_size:
                chunks = []
                for i in range(0, len(audio), max_chunk_size):
                    chunk = audio[i:i + max_chunk_size]
                    chunk_tensor = torch.FloatTensor(chunk).to(config.device)
                    chunk_opt = vc.pipeline(
                        hubert_model, net_g, 0, chunk_tensor, 
                        "chunk" if vc_input else "temp", times, 
                        int(f0_up_key), f0_method, file_index, index_rate,
                        if_f0, filter_radius, tgt_sr, resample_sr,
                        rms_mix_rate, version, protect, f0_file=None,
                    )
                    chunks.append(chunk_opt)
                audio_opt = np.concatenate(chunks)
            else:
                audio_opt = vc.pipeline(
                    hubert_model, net_g, 0, audio_tensor, 
                    vc_input if vc_input else "temp", times, 
                    int(f0_up_key), f0_method, file_index, index_rate,
                    if_f0, filter_radius, tgt_sr, resample_sr,
                    rms_mix_rate, version, protect, f0_file=None,
                )
            
            audio_opt = audio_opt.astype(np.float32)
            if speed != 1.0:
                audio_opt = adjust_audio_speed(audio_opt, speed)
            if np.max(np.abs(audio_opt)) > 0:
                audio_opt = (audio_opt / np.max(np.abs(audio_opt)) * 0.9).astype(np.float32)
            
            yield "Status: ✅ Selesai!", (tgt_sr, audio_opt)
            
        except Exception as e:
            yield f"❌ Error: {str(e)}\n\n{traceback.format_exc()}", None
        finally:
            if temp_audio_file and os.path.exists(temp_audio_file):
                os.remove(temp_audio_file)
            if torch.cuda.is_available():
                torch.cuda.empty_cache()
            if model_key not in model_cache:
                net_g.to('cpu')
    
    return vc_fn

def create_model_info_from_files(base_path):
    mygo_dir = os.path.join(base_path, "BanGDream-MyGO")
    if not os.path.exists(mygo_dir):
        return
    
    model_info_path = os.path.join(mygo_dir, "model_info.json")
    model_info = {}
    
    for char_folder in os.listdir(mygo_dir):
        char_path = os.path.join(mygo_dir, char_folder)
        if not os.path.isdir(char_path):
            continue
        pth_files = [f for f in os.listdir(char_path) if f.endswith('.pth')]
        index_files = [f for f in os.listdir(char_path) if f.endswith('.index')]
        image_files = [f for f in os.listdir(char_path) if f.lower().endswith(('.png', '.jpg', '.jpeg'))]
        if not pth_files:
            continue
        char_name_formatted = re.sub(r"([a-z])([A-Z])", r"\1 \2", char_folder)
        model_info[char_folder] = {
            "enable": True,
            "model_path": pth_files[0],
            "title": f"MyGO - {char_name_formatted}",
            "cover": image_files[0] if image_files else "cover.png",
            "feature_retrieval_library": index_files[0] if index_files else "",
            "author": "Plana-Archive"
        }
    
    with open(model_info_path, "w", encoding="utf-8") as f:
        json.dump(model_info, f, indent=2, ensure_ascii=False)
    print(f"✅ Created model_info.json with {len(model_info)} characters")
    return model_info

def load_model():
    categories = []
    base_path = "weights"
    if not os.path.exists(base_path):
        print(f"❌ Folder '{base_path}' not found!")
        return categories
    
    folder_info_path = f"{base_path}/folder_info.json"
    if not os.path.isfile(folder_info_path):
        folder_info = {
            "BanGDream-MyGO": {
                "title": "BanG Dream! MyGO!!!!!",
                "folder_path": "BanGDream-MyGO",
                "description": "Official RVC Weights for BanG Dream! MyGO!!!!! characters",
                "enable": True
            }
        }
        with open(folder_info_path, "w", encoding="utf-8") as f:
            json.dump(folder_info, f, indent=2, ensure_ascii=False)
    
    with open(folder_info_path, "r", encoding="utf-8") as f:
        folder_info = json.load(f)
    
    for category_name, category_info in folder_info.items():
        if not category_info.get('enable', True): 
            continue
        category_title, category_folder, description = (
            category_info['title'], 
            category_info['folder_path'], 
            category_info['description']
        )
        models = []
        model_info_path = f"{base_path}/{category_folder}/model_info.json"
        if not os.path.exists(model_info_path):
            model_info = create_model_info_from_files(base_path)
            if not model_info:
                continue
        
        if os.path.exists(model_info_path):
            with open(model_info_path, "r", encoding="utf-8") as f:
                models_info = json.load(f)
            for character_name, info in models_info.items():
                if not info.get('enable', True): 
                    continue
                model_title, model_name, model_author = (
                    info['title'], 
                    info['model_path'], 
                    info.get("author")
                )
                cache_key = f"{category_folder}_{character_name}"
                if cache_key in model_cache:
                    tgt_sr, net_g, vc, if_f0, version, model_index = model_cache[cache_key]
                else:
                    model_cover = f"{base_path}/{category_folder}/{character_name}/{info['cover']}"
                    model_index = f"{base_path}/{category_folder}/{character_name}/{info['feature_retrieval_library']}"
                    model_path = f"{base_path}/{category_folder}/{character_name}/{model_name}"
                    cpt = torch.load(model_path, map_location="cpu")
                    tgt_sr = cpt["config"][-1]
                    cpt["config"][-3] = cpt["weight"]["emb_g.weight"].shape[0]
                    if_f0, version = cpt.get("f0", 1), cpt.get("version", "v1")
                    if version == "v1":
                        if if_f0 == 1:
                            net_g = SynthesizerTrnMs256NSFsid(*cpt["config"], is_half=config.is_half)
                        else:
                            net_g = SynthesizerTrnMs256NSFsid_nono(*cpt["config"])
                    else:
                        if if_f0 == 1:
                            net_g = SynthesizerTrnMs768NSFsid(*cpt["config"], is_half=config.is_half)
                        else:
                            net_g = SynthesizerTrnMs768NSFsid_nono(*cpt["config"])
                    if hasattr(net_g, "enc_q"): 
                        del net_g.enc_q
                    net_g.load_state_dict(cpt["weight"], strict=False)
                    net_g.eval().to('cpu')
                    vc = VC(tgt_sr, config)
                    model_cache[cache_key] = (tgt_sr, net_g, vc, if_f0, version, model_index)
                models.append((
                    character_name, model_title, model_author, 
                    f"{base_path}/{category_folder}/{character_name}/{info['cover']}", 
                    version, 
                    create_vc_fn(cache_key, tgt_sr, net_g, vc, if_f0, version, model_index)
                ))
        categories.append([category_title, category_folder, description, models])
    return categories

def load_hubert():
    global hubert_model, hubert_loaded
    if hubert_loaded:
        return
    torch.serialization.add_safe_globals([Dictionary])
    models, _, _ = checkpoint_utils.load_model_ensemble_and_task(
        ["hubert_base.pt"], 
        suffix="",
    )
    hubert_model = models[0].to(config.device)
    hubert_model = hubert_model.half() if config.is_half else hubert_model.float()
    hubert_model.eval()
    hubert_loaded = True

def change_audio_mode(vc_audio_mode):
    is_input_path = vc_audio_mode == "Input path"
    is_upload = vc_audio_mode == "Upload audio"
    is_tts = vc_audio_mode == "TTS Audio"
    
    return (
        gr.Textbox.update(visible=is_input_path),
        gr.Audio.update(visible=is_upload),
        gr.Textbox.update(visible=is_tts, lines=4 if is_tts else 2)
    )

# CSS (sama seperti sebelumnya)
css = """
@import url('https://fonts.googleapis.com/css2?family=Inter:wght@300;400;500;600;700&family=Quicksand:wght@400;600;700&display=swap');
body, .gradio-container { background-color: #ffffff !important; font-family: 'Inter', sans-serif !important; }
footer { display: none !important; }
.arona-loading-container { display: flex; align-items: center; justify-content: center; gap: 15px; margin-top: 15px; padding: 10px; }
.loading-text-blue { font-family: 'Quicksand', sans-serif; font-size: 20px; font-weight: 700; color: #00b0ff; letter-spacing: 1px; }
.loading-gif-small { width: 100px; height: auto; border-radius: 8px; }
.header-img-container { text-align: center; padding: 10px 0; background: #ffffff !important; }
.header-img { width: 100%; max-width: 500px; border-radius: 15px; margin: 0 auto; display: block; }
.status-card { background: #ffffff; border: 1px solid #e1f0ff; border-radius: 14px; padding: 15px 10px; margin: 0 auto 15px auto; max-width: 400px; display: flex; flex-direction: column; align-items: center; }
.status-online-box { display: flex; align-items: center; gap: 8px; margin-bottom: 12px; }
.status-details-container { display: flex; width: 100%; justify-content: center; align-items: center; border-top: 1px solid #f0f7ff; padding-top: 10px; }
.status-detail-item { flex: 1; display: flex; flex-direction: column; align-items: center; text-align: center; }
.status-detail-item:first-child { border-right: 1px solid #e1f0ff; }
.status-text-main { font-size: 13px !important; font-weight: 600; color: #546e7a; }
.status-text-sub { font-size: 11px !important; color: #90a4ae; }
.dot-online { height: 8px; width: 8px; background-color: #2ecc71; border-radius: 50%; display: inline-block; animation: blink-green 1.5s infinite; }
@keyframes blink-green { 0% { opacity: 1; } 50% { opacity: 0.4; } 100% { opacity: 1; } }
.gr-form .gr-block label span, .gr-box label span, .gr-panel label span { background: linear-gradient(135deg, #4fc3f7 0%, #00b0ff 100%) !important; color: white !important; padding: 4px 12px !important; border-radius: 8px !important; font-weight: 600 !important; box-shadow: 0 0 15px rgba(79, 195, 247, 0.4) !important; }
input[type="range"] { accent-color: #00b0ff !important; }
.char-scroll-box { display: grid !important; grid-template-columns: repeat(2, 1fr) !important; gap: 12px !important; max-height: 280px; overflow-y: auto; padding: 15px; background: #ffffff; border: 2px solid #eef5ff; border-radius: 14px; }
.char-card { background: white; padding: 12px; border-radius: 12px; cursor: pointer; border: 1px solid #e1f5fe; border-left: 5px solid #4fc3f7; transition: all 0.2s ease; display: flex; flex-direction: column; height: 65px; }
.char-name-jp { font-weight: 700; font-size: 11px !important; color: #455a64; }
.char-name-en { font-size: 8.5px !important; color: #90a4ae; text-transform: uppercase; }
.speed-section { margin-top: 20px; padding: 18px; border-radius: 20px; background: linear-gradient(135deg, #f0f7ff 0%, #ffffff 100%); border: 2px solid #e1f0ff; }
.speed-title { font-family: 'Quicksand', sans-serif; font-weight: 700; color: #4ea8de; text-align: center; margin-bottom: 12px; font-size: 14px; }
.generate-btn { font-family: 'Quicksand', sans-serif; font-weight: 700 !important; background: linear-gradient(135deg, #64b5f6 0%, #2196f3 100%) !important; color: white !important; border-radius: 12px !important; }
.footer-text { text-align: center; padding: 20px; border-top: 1px solid #f0f4f8; color: #b0bec5; font-size: 11px; }
.speed-notes-box { font-family: 'Arial'; border: 1px solid #ffd8b2; border-radius: 8px; padding: 12px; background: #fff7ed; border-left: 4px solid #fb923c; margin-top: 10px; }
.speed-notes-title { color: #c2410c; font-size: 12px; margin: 0 0 5px 0; font-weight: bold; }
.speed-notes-content { color: #9a3412; font-size: 11px; margin: 0; }
.video-demo-container { text-align: center; padding: 20px; background: #ffffff; border-radius: 20px; border: 2px solid #e1f0ff; margin: 20px auto; max-width: 800px; }
.video-demo-title { font-family: 'Quicksand', sans-serif; font-weight: 700; color: #4fc3f7; font-size: 18px; margin-bottom: 15px; }
.video-demo-player { width: 100%; border-radius: 15px; box-shadow: 0 10px 30px rgba(0, 176, 255, 0.2); }
"""

if __name__ == '__main__':
    load_hubert()
    categories = load_model()
    total_models = sum(len(models) for _, _, _, models in categories)
    
    with gr.Blocks(css=css, theme=gr.themes.Soft()) as app:
        gr.HTML('<div class="header-img-container"><img src="https://huggingface.co/spaces/BanG-Dream-MyGO/BanG-Dream-MyGO/resolve/main/MyGO.PNG" class="header-img"></div>')
        
        gr.HTML(f'''
            <div class="status-card">
                <div class="status-online-box"><span class="dot-online"></span><b style="color: #4fc3f7; font-size: 14px;">System Online</b></div>
                <div class="status-details-container">
                    <div class="status-detail-item"><span class="status-text-main">🎸 {total_models} Members</span><span class="status-text-sub">Ready</span></div>
                    <div class="status-detail-item"><span class="status-text-main">📊 Total</span><span class="status-text-sub">Database: {total_models}</span></div>
                </div>
            </div>
        ''')
        
        with gr.Row():
            with gr.Column(scale=1):
                pass
            with gr.Column(scale=3):
                gr.HTML("""
                    <div class="video-demo-container">
                        <div class="video-demo-title">✅ [ON] MODE YURI 💚</div>
                        <video class="video-demo-player" controls autoplay loop muted playsinline>
                            <source src="https://huggingface.co/spaces/BanG-Dream-MyGO/README/resolve/main/Sakiko Chan.mp4" type="video/mp4">
                            Your browser does not support the video tag.
                        </video>
                    </div>
                """)
            with gr.Column(scale=1):
                pass
        
        for cat_idx, (folder_title, folder, description, models) in enumerate(categories):
            with gr.TabItem(folder_title):
                with gr.Accordion("📑 Select Member", open=True):
                    char_html = "".join([f'<div class="char-card" onclick="selectModel(\'{folder_title}\', \'{name}\')"><span class="char-name-jp">{clean_title(title)}</span><span class="char-name-en">{name}</span></div>' for name, title, author, cover, version, vc_fn in models])
                    gr.HTML(f'<div class="char-scroll-box">{char_html}</div>')
                
                with gr.Tabs():
                    for model_idx, (name, title, author, cover, model_version, vc_fn) in enumerate(models):
                        with gr.TabItem(name, id=f"model_{cat_idx}_{model_idx}"):
                            with gr.Row():
                                with gr.Column(scale=1):
                                    gr.HTML(f'<div style="display:flex;flex-direction:column;align-items:center;padding:20px;background:white;border-radius:20px;border:1px solid #eef5ff;"><img style="width:200px;height:260px;object-fit:cover;border-radius:15px;" src="file/{cover}"><div style="font-family:\'Quicksand\',sans-serif;font-weight:700;font-size:18px;color:#039be5;margin-top:15px;">{clean_title(title)}</div><div style="font-size:11px;color:#b0bec5;margin-top:5px;">{model_version} • {author}</div></div>')
                                
                                with gr.Column(scale=2):
                                    with gr.Group():
                                        vc_audio_mode = gr.Dropdown(label="Input Mode", choices=audio_mode, value="TTS Audio")
                                        vc_input = gr.Textbox(visible=False)
                                        # HAPUS CHECKBOX MICROPHONE – langsung gunakan upload audio
                                        vc_upload = gr.Audio(label="Upload Audio Source", visible=False)
                                        tts_text = gr.Textbox(label="TTS Text", visible=True, placeholder="Type message here...", lines=3)
                                    
                                    with gr.Row():
                                        with gr.Column():
                                            vc_transform0 = gr.Slider(minimum=-12, maximum=12, label="Pitch (Nada)", value=12, step=1)
                                            f0method0 = gr.Radio(label="Conversion Algorithm", choices=f0method_mode, value="rmvpe")
                                        with gr.Column():
                                            with gr.Accordion("⚙️ SETTINGS ⚙️", open=False):
                                                index_rate1 = gr.Slider(0, 1, label="Index Rate", value=0.75)
                                                filter_radius0 = gr.Slider(0, 7, label="Filter", value=7, step=1)
                                                resample_sr0 = gr.Slider(0, 48000, label="Resample", value=0)
                                                rms_mix_rate0 = gr.Slider(0, 1, label="Volume Mix", value=0.76)
                                                protect0 = gr.Slider(0, 0.5, label="Voice Protect", value=0.33)
                                    
                                    with gr.Column(elem_classes="speed-section"):
                                        gr.HTML('<div class="speed-title">⚡ KECEPATAN SUARA ⚡</div>')
                                        speed_slider = gr.Slider(0.5, 2.0, value=1.0, step=0.1, label=None)
                                        gr.HTML("""<div class="speed-notes-box">
                                            <div class="speed-notes-title">🌥️ CATATAN KECIL 🌥️</div>
                                            <div class="speed-notes-content">
                                                • <b>Kiri (0.5):</b> untuk mempercepat Suara<br>
                                                • <b>✅ Tengah (1.0):</b> untuk Kecepatan normal (disarankan)<br>
                                                • <b>🚫 Kanan (2.0):</b> Mempercepat suara (tidak di sarankan)<br><br>
                                                <b>Tips:</b> Atur ke kiri untuk suara lebih lambat dan atur ke kanan untuk suara lebih cepat. Disarankan tetap di 1.0 untuk hasil normal atau ubah jadi 08 atau 09.
                                            </div>
                                        </div>""")
                                        gr.HTML('<div class="arona-loading-container"><div class="loading-text-blue">Let\'s Play Music!</div><img class="loading-gif-small" src="https://huggingface.co/spaces/BanG-Dream-MyGO/BanG-Dream-MyGO/resolve/main/Mutsumi-Chan.gif"></div>')
                                
                                with gr.Column(scale=1):
                                    vc_log = gr.Textbox(label="Process Logs", interactive=False)
                                    vc_output = gr.Audio(label="Result Audio", interactive=False)
                                    vc_convert = gr.Button("🎸 GENERATE VOICE 🎸", variant="primary", elem_classes="generate-btn")

                            vc_convert.click(
                                fn=vc_fn, 
                                inputs=[vc_audio_mode, vc_input, vc_upload, tts_text, vc_transform0, f0method0, 
                                       index_rate1, filter_radius0, resample_sr0, rms_mix_rate0, protect0, speed_slider], 
                                outputs=[vc_log, vc_output]
                            )
                            vc_audio_mode.change(fn=change_audio_mode, inputs=[vc_audio_mode], outputs=[vc_input, vc_upload, tts_text])

        gr.HTML('<div class="footer-text"><div>DESIGNED BY 🎸 Mutsumi Chan 🎸</div><div style="font-weight:700; color:#90a4ae;">BanG Dream! MyGO!!!!! Voice Conversion • 2024</div></div>')
        app.load(None, None, None, js="""() => { window.selectModel = (cat, mod) => { const tabs = document.querySelectorAll('.tabs .tab-nav button'); for (let t of tabs) { if (t.textContent.trim() === cat) { t.click(); setTimeout(() => { const mTabs = document.querySelectorAll('.tabs .tab-nav button'); for (let mt of mTabs) { if (mt.textContent.trim() === mod) mt.click(); } }, 50); break; } } } }""")

    # PERBAIKAN: Gunakan share=True dan tanpa parameter server_name/port
    app.queue(max_size=3).launch(share=True)