File size: 15,579 Bytes
d4b1959 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 | 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 shutil
from datetime import datetime
from fairseq import checkpoint_utils
from fairseq.data.dictionary import Dictionary
from huggingface_hub import snapshot_download
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)
# --- KONFIGURASI DOWNLOAD OTOMATIS DARI REPO MODEL (UPDATED) ---
# Mengakses repository baru berdasarkan struktur di screenshot
if not os.path.exists("weights"):
print("Mendownload weights dan bahan model dari repo Plana-RCV/BanGDream-MyGO...")
snapshot_download(
repo_id="Plana-Archive/Anime-RCV",
local_dir=".",
allow_patterns=[
"BanGDream-MyGO/weights/*",
"BanGDream-MyGO/hubert_base.pt",
"BanGDream-MyGO/rmvpe.pt"
],
repo_type="model"
)
source_dir = "BanGDream-MyGO"
if os.path.exists(source_dir):
print(f"Menyusun ulang struktur folder dari {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)
os.rmdir(source_dir)
print("Struktur folder berhasil diperbarui.")
spaces = True
# Setup metode F0
f0method_mode = ["pm", "harvest"]
if os.path.isfile("rmvpe.pt"):
f0method_mode.insert(2, "rmvpe")
def _load_audio_input(tts_text, speed, spaces_limit=20):
temp_file = "tts.mp3"
if not tts_text or tts_text.strip() == "":
return None, None, "EMPTY"
if len(tts_text) > 100 and spaces:
return None, None, "TOO_LONG"
speed_rate = f"{'+' if speed >= 1.0 else '-'}{int(abs(speed - 1.0) * 100)}%"
tts_voice_default = "ja-JP-NanamiNeural"
asyncio.run(edge_tts.Communicate(tts_text, tts_voice_default, rate=speed_rate).save(temp_file))
audio, sr = librosa.load(temp_file, sr=16000, mono=True)
return audio, sr, temp_file
def create_vc_fn(model_name, tgt_sr, net_g, vc, if_f0, version, file_index):
def vc_fn(
tts_text,
f0_up_key, f0_method, index_rate, filter_radius,
resample_sr, rms_mix_rate, protect, speed_rate,
):
logs = []
temp_audio_file = "tts.mp3"
try:
audio, sr, status = _load_audio_input(tts_text, speed_rate)
if status == "EMPTY":
return "⚠️ Mohon masukkan teks terlebih dahulu!", None
if status == "TOO_LONG":
return "❌ Teks terlalu panjang! Maksimal 100 karakter.", None
logs.append(f"✨ Model: {model_name}")
yield "\n".join(logs), None
logs.append("📥 Memuat audio dasar...")
logs.append(f"⚙️ Memproses RVC (Pitch: {f0_up_key})...")
yield "\n".join(logs), None
times = [0, 0, 0]
audio_opt = vc.pipeline(
hubert_model, net_g, 0, audio, status,
times, 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,
)
logs.append(f"✅ Selesai pada: {datetime.now().strftime('%H:%M:%S')}")
yield "\n".join(logs), (tgt_sr, audio_opt)
except Exception as e:
traceback.print_exc()
return f"❌ Error: {str(e)}", None
finally:
if os.path.exists(temp_audio_file):
os.remove(temp_audio_file)
return vc_fn
def load_model():
categories = []
folder_info_path = "weights/folder_info.json"
if os.path.isfile(folder_info_path):
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_info['title']
category_folder = category_info['folder_path']
models = []
model_info_path = os.path.join("weights", category_folder, "model_info.json")
if not os.path.exists(model_info_path): continue
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
base_character_path = os.path.join("weights", category_folder, character_name)
cpt_path = os.path.join(base_character_path, info['model_path'])
model_cover = os.path.join(base_character_path, info['cover'])
model_index = os.path.join(base_character_path, info['feature_retrieval_library'])
if not os.path.exists(cpt_path): continue
cpt = torch.load(cpt_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"])
elif version == "v2":
net_g = SynthesizerTrnMs768NSFsid(*cpt["config"], is_half=config.is_half) if if_f0 == 1 else 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(config.device)
net_g = net_g.half() if config.is_half else net_g.float()
vc = VC(tgt_sr, config)
models.append((character_name, info['title'], info.get("author"), model_cover, version, create_vc_fn(info['model_path'], tgt_sr, net_g, vc, if_f0, version, model_index)))
categories.append([category_title, category_folder, category_info.get('description',''), 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"], suffix="",)
hubert_model = models[0].to(config.device)
hubert_model = hubert_model.half() if config.is_half else hubert_model.float()
hubert_model.eval()
if __name__ == '__main__':
load_hubert()
categories = load_model()
total_characters = sum(len(cat[3]) for cat in categories)
# UPDATED: Background set to White (#ffffff)
custom_css = """
.gradio-container { background-color: #ffffff !important; }
.tabs { background-color: #ffffff !important; border-radius: 12px; border: 1px solid #d1f2d1 !important; }
.primary-btn { background-color: #a8e6cf !important; border: none !important; color: white !important; font-weight: bold !important; }
.primary-btn:hover { background-color: #89d9bb !important; }
"""
with gr.Blocks(theme=gr.themes.Soft(primary_hue="green", secondary_hue="emerald"), css=custom_css) as app:
gr.HTML(f"""
<div style="font-family: 'Arial', sans-serif; max-width: 800px; margin: 20px auto 10px auto; border: 1px solid #d1f2d1; border-radius: 12px; padding: 15px; background-color: #ffffff; text-align: center;">
<h1 style="color: #2d6a4f; margin: 0;">BanG Dream! RVC</h1>
<p style="color: #52b788; font-size: 14px; margin-top: 5px;">RVC-BANG-DREAM • Weights by Plana-Archive</p>
</div>
<div style="font-family: 'Arial', sans-serif; max-width: 800px; margin: 0 auto 20px auto; border: 1px solid #d1f2d1; border-radius: 10px; padding: 15px; background-color: white; display: flex; justify-content: space-around; align-items: center;">
<div style="text-align: center;">
<p style="color: #94a3b8; font-size: 11px; font-weight: 700; margin: 0; text-transform: uppercase;">System Status</p>
<p style="color: #22c55e; font-size: 14px; font-weight: 700; margin: 0;">● ONLINE</p>
</div>
<div style="height: 30px; border-left: 1px solid #f1f5f9;"></div>
<div style="text-align: center;">
<p style="color: #94a3b8; font-size: 11px; font-weight: 700; margin: 0; text-transform: uppercase;">Total Characters</p>
<p style="color: #1e293b; font-size: 14px; font-weight: 700; margin: 0;">{total_characters} Models</p>
</div>
</div>
""")
for (folder_title, folder, description, models) in categories:
with gr.TabItem(folder_title):
with gr.Tabs():
for (name, title, author, cover, model_version, vc_fn) in models:
with gr.TabItem(name):
with gr.Row():
gr.Markdown(f'<div align="center"><h3 style="color: #1b4332;">{title}</h3>' + (f'<img style="width:auto;height:250px;border-radius:10px;border: 3px solid #a8e6cf;" src="file/{cover}">' if cover else "") + '</div>')
with gr.Row():
with gr.Column():
tts_text = gr.Textbox(label="🏷️ MASUK TEXT SINI", info="Masukkan teks yang ingin diucapkan", lines=3)
vc_pitch = gr.Slider(minimum=-12, maximum=12, label="Pitch (Nada)", value=12, step=1, info="Diset ke +12 untuk karakter perempuan")
with gr.Column():
f0method0 = gr.Radio(label="Algoritma Pitch", choices=f0method_mode, value="rmvpe" if "rmvpe" in f0method_mode else "pm")
index_rate1 = gr.Slider(minimum=0, maximum=1, label="Rasio Retrieval", value=0.75)
filter_radius0 = gr.Slider(minimum=0, maximum=7, label="Median Filtering", value=7, step=1)
with gr.Column():
resample_sr0 = gr.Slider(minimum=0, maximum=48000, label="Resample Rate", value=0, step=1)
rms_mix_rate0 = gr.Slider(minimum=0, maximum=1, label="Volume Envelope", value=0.76)
protect0 = gr.Slider(minimum=0, maximum=0.5, label="Proteksi Suara", value=0.33, step=0.01)
gr.HTML("""<div style="font-family: 'Arial', sans-serif; border: 1px solid #bae6fd; border-radius: 10px; padding: 15px; background-color: #f0f9ff; margin-bottom: 10px; border-left: 5px solid #0ea5e9;"><h4 style="color: #0369a1; font-size: 14px; font-weight: 700; margin: 0 0 8px 0;">📝 Notes & Panduan Fitur 📑</h4><ul style="color: #075985; font-size: 12px; margin: 0; padding-left: 18px; line-height: 1.5;"><li><b>Algoritma Pitch:</b> Akurasi nada (RMVPE terbaik).</li><li><b>Rasio Retrieval:</b> Kemiripan karakter asli (0.7+).</li><li><b>Median Filtering:</b> Menghilangkan suara kresek/noise.</li><li><b>Resample Rate:</b> Kejernihan audio (0 otomatis).</li><li><b>Volume Envelope:</b> Keseimbangan volume suara.</li><li><b>Proteksi Suara:</b> Melindungi suara alami manusia.</li></ul></div>""")
gr.HTML("""<div style="font-family: 'Arial', sans-serif; border: 1px solid #dcfce7; border-radius: 10px; padding: 15px; background-color: #f0fdf4; margin-bottom: 10px; border-left: 5px solid #22c55e;"><h4 style="color: #166534; font-size: 14px; font-weight: 700; margin: 0 0 8px 0;">📑 DI SARANKAN 📑</h4><ul style="color: #166534; font-size: 11px; margin: 0; padding-left: 18px; line-height: 1.6;"><li><b>Algoritma Pitch:</b> Selalu gunakan <b>RMVPE</b> untuk kejernihan maksimal.</li><li><b>Rasio Retrieval:</b> Set di angka <b>0.75</b> untuk kemiripan karakter.</li><li><b>Median Filtering:</b> Gunakan angka <b>7</b> untuk suara paling bersih.</li><li><b>Resample Rate:</b> Set ke <b>0</b> (Otomatis) agar tidak pecah.</li><li><b>Volume Envelope:</b> Gunakan <b>0.76</b> untuk kestabilan suara.</li><li><b>Proteksi Suara:</b> Set ke <b>0.33</b> agar hasil tidak kaku/robotik.</li><li><b>Pitch:</b> Naikkan ke <b>+12</b> khusus untuk karakter perempuan.</li></ul></div>""")
speed_rate = gr.Slider(minimum=0.5, maximum=2.0, label="Kecepatan Suara", value=1.0, step=0.1)
gr.HTML("""<div style="margin-bottom: -15px;"><span style="color: #40916c; font-weight: 700; font-size: 13px;">🖥️ LOG SISTEM</span></div>""")
vc_log = gr.Textbox(label="", placeholder="Menunggu proses...", interactive=False)
vc_output = gr.Audio(label="Audio Hasil", interactive=False)
vc_convert = gr.Button("🎸 GENERATE VOICE 🎸", variant="primary", elem_classes="primary-btn")
# --- BAGIAN TAMBAHAN: PERINGATAN MINNA ---
gr.HTML("""
<div style="font-family: 'Arial', sans-serif; border: 1px solid #fecaca; border-radius: 10px; padding: 15px; background-color: #fef2f2; margin-top: 15px; border-left: 5px solid #ef4444;">
<h4 style="color: #991b1b; font-size: 14px; font-weight: 700; margin: 0 0 5px 0;">PERINGATAN MINNA 🔖</h4>
<p style="color: #b91c1c; font-size: 12px; margin: 0; line-height: 1.5;">
Setelah di Generate Voice, audionya akan muncul beberapa detik dan tunggu aja ya!
</p>
</div>
""")
vc_convert.click(
fn=vc_fn,
inputs=[tts_text, vc_pitch, f0method0, index_rate1, filter_radius0, resample_sr0, rms_mix_rate0, protect0, speed_rate],
outputs=[vc_log, vc_output]
)
gr.HTML("""<div style="font-family: 'Arial', sans-serif; max-width: 800px; margin: 30px auto 20px auto; border: 1px solid #d1f2d1; border-radius: 12px; padding: 20px; background-color: white; text-align: center;"><h3 style="color: #1b4332; font-size: 16px; margin: 0; font-weight: 700;">CREATED BY PLANA-CHAN</h3><p style="color: #94a3b8; font-size: 13px; margin-top: 4px;">BanG Dream! RVC Implementation</p></div>""")
app.queue(max_size=20).launch(share=False, server_name="0.0.0.0", server_port=7860) |