import os import glob import json import traceback import logging import gradio as gr import numpy as np import librosa import torch import asyncio import edge_tts import sys import io import wave import re import shutil import time from datetime import datetime from huggingface_hub import snapshot_download 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 config = Config() logging.getLogger("numba").setLevel(logging.WARNING) logging.getLogger("fairseq").setLevel(logging.WARNING) # === SISTEM AUTO-DOWNLOAD === def download_required_weights(): repo_id = "Plana-Archive/Anime-RCV" # First download folder_info.json print("πŸ“¦ Downloading folder_info.json...") try: snapshot_download( repo_id=repo_id, allow_patterns=["Waifu-Anime-RCV/weights/folder_info.json"], local_dir=".", local_dir_use_symlinks=False ) print("βœ… folder_info.json downloaded!") except Exception as e: print(f"⚠️ Failed to download folder_info.json: {e}") return # Load folder_info.json folder_info_path = "Waifu-Anime-RCV/weights/folder_info.json" if not os.path.exists(folder_info_path): print("❌ folder_info.json not found!") return with open(folder_info_path, "r", encoding="utf-8") as f: folder_info = json.load(f) # Download each enabled category for cat_key, cat_info in folder_info.items(): if cat_info.get("enable", False): folder_path = cat_info["folder_path"] print(f"πŸ“¦ Downloading models from {folder_path}...") try: snapshot_download( repo_id=repo_id, allow_patterns=[f"{folder_path}/*"], local_dir=".", local_dir_use_symlinks=False ) print(f"βœ… {folder_path} downloaded!") except Exception as e: print(f"⚠️ Failed to download {folder_path}: {e}") download_required_weights() 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'^Blue Archive\s*-\s*', '', title, flags=re.IGNORECASE) title = re.sub(r'^Waifu Lovers\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, use_mic, mic_input): temp_file = None try: if use_mic and mic_input is not None: sampling_rate, audio = mic_input 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) return audio.astype(np.float32), 16000, None 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("Upload audio!") 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) return audio.astype(np.float32), 16000, None elif vc_audio_mode == "TTS Audio": if not tts_text: raise ValueError("Isi teks!") temp_file = "tts_temp.wav" async def tts_task(): return await edge_tts.Communicate(tts_text, "ja-JP-NanamiNeural").save(temp_file) asyncio.run(tts_task()) audio, sr = librosa.load(temp_file, sr=16000, mono=True) return audio.astype(np.float32), 16000, temp_file except Exception as e: raise e 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, use_mic, mic_input, f0_up_key, f0_method, index_rate, filter_radius, resample_sr, rms_mix_rate, protect, speed): temp_audio_file = None try: yield "Status: πŸš€ Memproses...", None audio, sr, temp_audio_file = _load_audio_input(vc_audio_mode, vc_input, vc_upload, tts_text, use_mic, mic_input) audio_tensor = torch.FloatTensor(audio).to(config.device) times = [0, 0, 0] audio_opt = vc.pipeline(hubert_model, net_g, 0, audio_tensor, "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) if speed != 1.0: audio_opt = librosa.effects.time_stretch(audio_opt.astype(np.float32), rate=speed) yield "Status: βœ… Selesai!", (tgt_sr, audio_opt) except Exception as e: yield f"❌ Error: {str(e)}", None finally: if temp_audio_file and os.path.exists(temp_audio_file): os.remove(temp_audio_file) return vc_fn def load_model(): categories = [] # Load folder_info.json folder_info_path = "Waifu-Anime-RCV/weights/folder_info.json" if not os.path.exists(folder_info_path): print("❌ folder_info.json not found!") return categories with open(folder_info_path, "r", encoding="utf-8") as f: folder_info = json.load(f) # Mapping untuk nama pendek kategori category_short_names = { "WaifuLovers": "🩷Waifu Lovers", "BanGDream": "🩷BanG Dream!", "Bocchi-the-Rock": "🩷Bocchi the Rock!", "Oshi-no-Ko": "🩷Oshi no Ko", "honkai-impact-3": "🩷Honkai Impact 3rd" } # Process each enabled category for cat_key, cat_info in folder_info.items(): if not cat_info.get("enable", False): continue folder_path = cat_info["folder_path"] category_title = cat_info["title"] category_description = cat_info.get("description", "") # Gunakan nama pendek jika ada, jika tidak gunakan yang asli display_title = category_short_names.get(cat_key, category_title) # Extract folder name from path base_path = os.path.dirname(folder_path) target_folder = os.path.basename(folder_path) models = [] # Check if category folder exists full_category_path = os.path.join(base_path, target_folder) if not os.path.exists(full_category_path): print(f"⚠️ Category folder not found: {full_category_path}") continue # Scan for character folders for char_name in os.listdir(full_category_path): char_dir = os.path.join(full_category_path, char_name) if not os.path.isdir(char_dir): continue model_info_path = os.path.join(char_dir, "model_info.json") if not os.path.exists(model_info_path): # Check for any .pth file as model pth_files = glob.glob(os.path.join(char_dir, "*.pth")) if pth_files: # Create a basic model_info entry model_path = pth_files[0] index_files = glob.glob(os.path.join(char_dir, "*.index")) index_path = index_files[0] if index_files else "" cover_files = glob.glob(os.path.join(char_dir, "*.png")) + glob.glob(os.path.join(char_dir, "*.jpg")) cover_path = cover_files[0] if cover_files else "" info = { 'title': char_name, 'model_path': os.path.basename(model_path), 'feature_retrieval_library': os.path.basename(index_path) if index_path else "", 'cover': os.path.basename(cover_path) if cover_path else "", 'author': 'Plana-Archive' } else: continue else: with open(model_info_path, "r", encoding="utf-8") as f: info = json.load(f) model_path = os.path.join(char_dir, info.get('model_path', '')) index_path = os.path.join(char_dir, info.get('feature_retrieval_library', '')) if os.path.exists(model_path): try: cpt = torch.load(model_path, map_location="cpu") tgt_sr, if_f0, version = cpt["config"][-1], cpt.get("f0", 1), cpt.get("version", "v1") if version == "v1": net_g = SynthesizerTrnMs256NSFsid(*cpt["config"], is_half=config.is_half) if if_f0==1 else SynthesizerTrnMs256NSFsid_nono(*cpt["config"]) else: net_g = SynthesizerTrnMs768NSFsid(*cpt["config"], is_half=config.is_half) if if_f0==1 else SynthesizerTrnMs768NSFsid_nono(*cpt["config"]) net_g.load_state_dict(cpt["weight"], strict=False) net_g.eval().to(config.device) vc = VC(tgt_sr, config) cover_path = os.path.join(char_dir, info.get('cover', '')) models.append(( char_name, info.get('title', char_name), info.get('author', 'Plana-Archive'), cover_path if os.path.exists(cover_path) else "", version, create_vc_fn(char_name, tgt_sr, net_g, vc, if_f0, version, index_path if os.path.exists(index_path) else None) )) print(f"βœ… Loaded model: {char_name} from {display_title}") except Exception as e: print(f"⚠️ Failed to load model {char_name}: {e}") if models: categories.append([display_title, target_folder, category_description, models]) print(f"βœ… Loaded category: {display_title} with {len(models)} models") return categories def load_hubert(): global hubert_model torch.serialization.add_safe_globals([Dictionary]) models, _, _ = checkpoint_utils.load_model_ensemble_and_task(["hubert_base.pt"]) hubert_model = models[0].to(config.device) hubert_model = hubert_model.half() if config.is_half else hubert_model.float() hubert_model.eval() # ============================= # FUNGSI TAMBAHAN UNTUK UI # ============================= def change_audio_mode(vc_audio_mode): """Mengubah tampilan input audio""" 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.Checkbox.update(visible=is_upload), gr.Audio.update(visible=is_upload), gr.Textbox.update(visible=is_tts, lines=4 if is_tts else 2) ) def use_microphone(microphone): """Toggle microphone/upload source""" return gr.Audio.update(source="microphone" if microphone else "upload") # ============================= # TAMPILAN MY WIFE 🩷 # ============================= 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-pink { font-family: 'Quicksand', sans-serif; font-size: 20px; font-weight: 700; color: #ff69b4; 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 #ffe4ec; 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 #fff0f7; 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 #ffe4ec; } .status-text-main { font-size: 13px !important; font-weight: 600; color: #7b4d5a; } .status-text-sub { font-size: 11px !important; color: #b07d8b; } .dot-online { height: 8px; width: 8px; background-color: #ff69b4; border-radius: 50%; display: inline-block; animation: blink-pink 1.5s infinite; } @keyframes blink-pink { 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, #ff69b4 0%, #ff1493 100%) !important; color: white !important; padding: 4px 12px !important; border-radius: 8px !important; font-weight: 600 !important; box-shadow: 0 0 15px rgba(255, 105, 180, 0.4) !important; } input[type="range"] { accent-color: #ff69b4 !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 #ffeef4; border-radius: 14px; } .char-card { background: white; padding: 12px; border-radius: 12px; cursor: pointer; border: 1px solid #ffe4ec; border-left: 5px solid #ff69b4; transition: all 0.2s ease; display: flex; flex-direction: column; height: 65px; } .char-card:hover { transform: translateY(-3px); box-shadow: 0 5px 15px rgba(255, 105, 180, 0.2); border-left-color: #ff1493; } .char-name-jp { font-weight: 700; font-size: 11px !important; color: #7b4d5a; } .char-name-en { font-size: 8.5px !important; color: #b07d8b; text-transform: uppercase; } .speed-section { margin-top: 20px; padding: 18px; border-radius: 20px; background: linear-gradient(135deg, #fff0f7 0%, #ffffff 100%); border: 2px solid #ffe4ec; } .speed-title { font-family: 'Quicksand', sans-serif; font-weight: 700; color: #ff69b4; text-align: center; margin-bottom: 12px; font-size: 14px; } .generate-btn { font-family: 'Quicksand', sans-serif; font-weight: 700 !important; background: linear-gradient(135deg, #ff69b4 0%, #ff1493 100%) !important; color: white !important; border-radius: 12px !important; padding: 12px 24px !important; transition: all 0.3s ease !important; } .generate-btn:hover { transform: scale(1.05); box-shadow: 0 5px 20px rgba(255, 20, 147, 0.3) !important; } .footer-text { text-align: center; padding: 20px; border-top: 1px solid #f8f0f4; color: #b07d8b; font-size: 11px; } .speed-notes-box { font-family: 'Arial'; border: 1px solid #ffd1dc; border-radius: 8px; padding: 12px; background: #fff5f8; border-left: 4px solid #ff69b4; margin-top: 10px; } .speed-notes-title { color: #ff1493; font-size: 12px; margin: 0 0 5px 0; font-weight: bold; } .speed-notes-content { color: #d81b60; font-size: 11px; margin: 0; } .model-tab { background: linear-gradient(135deg, #fff8fb 0%, #ffffff 100%) !important; border-radius: 15px !important; padding: 15px !important; } .advanced-settings { background: #f9f9f9 !important; border-radius: 10px !important; padding: 15px !important; border: 1px solid #e0e0e0 !important; } .error-box { background: #ffebee; border: 1px solid #ffcdd2; border-radius=8px; padding: 15px; margin: 10px 0; color: #c62828; } .info-box { background: #fce4ec; border: 1px solid #f8bbd9; border-radius=8px; padding: 15px; margin: 10px 0; color: #ad1457; } .peringatan-box { font-family: 'Arial'; border: 1px solid #ffd6e7; border-radius: 8px; padding: 15px; background: #fff0f7; border-left: 4px solid #ff69b4; margin-top: 20px; } .peringatan-title { color: #ff1493; font-size: 14px; margin: 0 0 8px 0; font-weight: bold; text-align: center; } .peringatan-content { color: #d81b60; font-size: 12px; margin: 0; text-align: center; font-weight: 600; } /* CSS untuk tab kategori yang lebih kompak */ .tabs { font-size: 14px !important; } .tab-nav { display: flex !important; flex-wrap: wrap !important; gap: 5px !important; margin-bottom: 10px !important; } .tab-nav button { font-size: 12px !important; padding: 6px 10px !important; margin: 2px !important; border-radius: 8px !important; background: linear-gradient(135deg, #ffe4ec 0%, #ffffff 100%) !important; border: 1px solid #ffb6c1 !important; color: #ff1493 !important; font-weight: 600 !important; transition: all 0.3s ease !important; min-width: 100px !important; text-align: center !important; } .tab-nav button:hover { background: linear-gradient(135deg, #ff69b4 0%, #ff1493 100%) !important; color: white !important; transform: translateY(-2px) !important; box-shadow: 0 3px 10px rgba(255, 105, 180, 0.3) !important; } .tab-nav button.selected { background: linear-gradient(135deg, #ff69b4 0%, #ff1493 100%) !important; color: white !important; border: 1px solid #ff1493 !important; } .category-tab-title { font-size: 12px !important; font-weight: 700 !important; text-align: center !important; white-space: nowrap !important; overflow: hidden !important; text-overflow: ellipsis !important; } /* CSS khusus untuk tabs model di dalam kategori */ .model-inner-tabs .tab-nav { display: flex !important; flex-wrap: wrap !important; gap: 3px !important; margin-bottom: 15px !important; } .model-inner-tabs .tab-nav button { font-size: 11px !important; padding: 5px 8px !important; margin: 1px !important; border-radius: 6px !important; background: linear-gradient(135deg, #f0f8ff 0%, #ffffff 100%) !important; border: 1px solid #d1e7ff !important; color: #4169e1 !important; font-weight: 600 !important; transition: all 0.3s ease !important; min-width: 80px !important; } .model-inner-tabs .tab-nav button:hover { background: linear-gradient(135deg, #4169e1 0%, #6495ed 100%) !important; color: white !important; transform: translateY(-1px) !important; } .model-inner-tabs .tab-nav button.selected { background: linear-gradient(135deg, #4169e1 0%, #6495ed 100%) !important; color: white !important; border: 1px solid #4169e1 !important; } /* CSS untuk video */ .video-demo-container { text-align: center; margin: 30px auto; padding: 20px; background: linear-gradient(135deg, #fff8fb 0%, #ffffff 100%); border-radius: 20px; border: 2px solid #ffe4ec; max-width: 600px; width: 100%; } .video-demo-title { font-family: 'Quicksand', sans-serif; font-weight: 700; color: #ff1493; font-size: 18px; margin-bottom: 15px; text-align: center; } .video-demo-player { border-radius: 15px; border: 2px solid #ffb6c1; width: 100%; max-width: 560px; height: auto; display: block; margin: 0 auto; } """ if __name__ == '__main__': load_hubert() categories = load_model() total_models = sum(len(m) for _, _, _, m in categories) with gr.Blocks(css=css, theme=gr.themes.Soft(primary_hue="pink")) as app: # HEADER DENGAN GAMBAR gr.HTML('
') # STATUS CARD DENGAN INFORMASI DOWNLOAD if total_models > 0: gr.HTML(f'''
Waifu Anime RCV System Online
πŸ‘₯ {total_models} Waifus Ready for Conversion
πŸ“Š Total Models Database: {total_models}
πŸ“₯ Model download completed! Ready to generate voices.
''') else: gr.HTML(f'''

⚠️ No Models Loaded

Please check console logs for details.

''') # TABS UNTUK SETIAP KATEGORI if categories: for cat_idx, (folder_title, folder, description, models) in enumerate(categories): with gr.TabItem(folder_title, elem_classes="model-tab"): with gr.Accordion("πŸ“‘ Character Information πŸ“‘", open=True): char_html = "".join([ f'
' f'{clean_title(title)}' f'{name}' f'
' for name, title, author, cover, version, vc_fn in models ]) gr.HTML(f'
{char_html}
') # TABS UNTUK SETIAP MODEL with gr.Tabs(elem_classes="model-inner-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(): # KOLOM KIRI: INFO MODEL with gr.Column(scale=1): if cover and os.path.exists(cover): gr.HTML(f'''
{clean_title(title)}
{model_version} β€’ {author}
''') else: gr.HTML(f'''
No Image
{clean_title(title)}
{model_version} β€’ {author}
''') # KOLOM TENGAH: INPUT DAN SETTINGS with gr.Column(scale=2): # INPUT GROUP with gr.Group(): vc_audio_mode = gr.Dropdown( label="Input Mode", choices=audio_mode, value="TTS Audio" ) vc_input = gr.Textbox(visible=False) vc_microphone_mode = gr.Checkbox( label="Use Microphone", value=False ) vc_upload = gr.Audio( label="Upload Audio Source", source="upload", visible=False, type="numpy" ) tts_text = gr.Textbox( label="TTS Text", visible=True, placeholder="Type your message here...", lines=4 ) # BASIC SETTINGS with gr.Row(): with gr.Column(): vc_pitch = gr.Slider( minimum=-12, maximum=12, label="Pitch", value=12, step=1 ) f0method0 = gr.Radio( label="Conversion Algorithm", choices=f0method_mode, value="rmvpe" if "rmvpe" in f0method_mode else "pm" ) with gr.Column(): with gr.Accordion("βš™οΈ Advanced Settings βš™οΈ", open=True, elem_classes="advanced-settings"): index_rate1 = gr.Slider( 0, 1, label="Index Rate", value=0.75 ) filter_radius0 = gr.Slider( 0, 7, label="Filter Radius", value=7, step=1 ) resample_sr0 = gr.Slider( 0, 48000, label="Resample SR", 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 ) # NOTES with gr.Row(): with gr.Column(): gr.HTML("""

πŸ“ Notes & Guide

Pitch: Adjust voice pitch

Algorithm: F0 extraction method

Retrieval: Voice similarity (0-1)

Filter: Noise reduction

Volume: Volume stability

Protect: Protect voice

""") with gr.Column(): gr.HTML("""

πŸ“‘ RECOMMENDED

Pitch: Untuk cewek ubah jadi (+12)

Pitch: Untuk Cowok ubah jadi (0)

Algorithm: RMVPE

Retrieval: 0.75

Filter: 7

Volume: 0.76

Protect: 0.33

""") # SPEED SECTION with gr.Column(elem_classes="speed-section"): gr.HTML('
⚑ VOICE SPEED CONTROL ⚑
') speed_slider = gr.Slider( 0.5, 2.0, value=1.0, step=0.1, label="Speed" ) gr.HTML("""
⚜️ Speed Voice ⚜️
β€’ Left (0.5): Slow down voice
β€’ Center (1.0): Normal speed
β€’ Right (2.0): Speed up voice
""") # LOADING INDICATOR gr.HTML( '
' '
Ready to Generate!
' '' '
' ) # KOLOM KANAN: OUTPUT with gr.Column(scale=1): vc_log = gr.Textbox( label="Process Logs", interactive=False, lines=4 ) vc_output = gr.Audio( label="Result Audio", interactive=False, type="numpy" ) vc_convert = gr.Button( "🩷 GENERATE VOICE 🩷", variant="primary", elem_classes="generate-btn", size="lg" ) # FITUR PERTAHANAN BAWAH GENERATE VOICE gr.HTML(f'''
πŸ”– PERINGATAN MINNA πŸ”–
Setelah di Generate Voice, audionya akan muncul beberapa detik dan tunggu aja ya!
βœ… (ON) MODE YURI - SAKI πŸ’š
""") with gr.Column(scale=1): pass # FOOTER gr.HTML( '' ) # JAVASCRIPT UNTUK MODEL SELECTION 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(); window.scrollTo({top: 0, behavior: 'smooth'}); } } }, 100); break; } } } } """ ) app.queue(max_size=10).launch(server_name="0.0.0.0", server_port=7860)