SPACE-RANDOM/RANDOM-SPACES/Blue-Archive-TTS-v4.0/app.py

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
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

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

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 "⚠️ Tolong isi teks dulu, Sensei! Kotak input tidak boleh kosong.", 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 = []
    if os.path.isfile("weights/folder_info.json"):
        with open("weights/folder_info.json", "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 = category_info['title'], category_info['folder_path']
            models = []
            
            model_info_path = f"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
                model_title, model_name = info['title'], info['model_path']
                model_cover = f"weights/{category_folder}/{character_name}/{info['cover']}"
                model_index = f"weights/{category_folder}/{character_name}/{info['feature_retrieval_library']}"
                
                cpt_path = f"weights/{category_folder}/{character_name}/{model_name}"
                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, model_title, info.get("author"), model_cover, version, create_vc_fn(model_name, 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)

    with gr.Blocks(theme=gr.themes.Soft()) as app:
        # Header (Logo & Stats)
        gr.HTML(f"""
            <div style="font-family: 'Arial', sans-serif; max-width: 800px; margin: 20px auto 10px auto; border: 1px solid #e0e6ed; border-radius: 12px; padding: 15px; background-color: #ffffff; text-align: center;">
                <img src="https://huggingface.co/spaces/Rosmontis-Chan/Blue-Archive-TTS/resolve/main/BLUE%20ARCHIVE.PNG" style="width: 100%; max-height: 220px; object-fit: contain; margin: 0 auto; display: block;" alt="Blue Archive Logo">
            </div>
            
            <div style="font-family: 'Arial', sans-serif; max-width: 800px; margin: 0 auto 20px auto; border: 1px solid #e0e6ed; 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: #1a9fff; 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 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;">Version</p>
                    <p style="color: #1e293b; font-size: 14px; font-weight: 700; margin: 0;">v4.0 Premium</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>{title}</h3>' + (f'<img style="width:auto;height:250px;border-radius:10px;" 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)
                                    # PITCH OTOMATIS +12
                                    vc_pitch = gr.Slider(minimum=-12, maximum=12, label="Pitch (Nada)", value=12, step=1, info="Otomatis diset ke +12 untuk karakter perempuan")
                                
                                with gr.Column():
                                    gr.HTML(f"""<div style="font-family: 'Arial', sans-serif; border: 1px solid #e0e6ed; border-radius: 10px; padding: 12px; background-color: white; margin-bottom: 10px;"><div style="display: flex; justify-content: space-between; align-items: center; cursor: pointer;" onclick="var x = document.getElementById('char_list_{name}'); var i = document.getElementById('icon_{name}'); if (x.style.display === 'none') {{ x.style.display = 'grid'; i.innerText = '▼'; }} else {{ x.style.display = 'none'; i.innerText = '«'; }}"><h4 style="color: #1a9fff; font-size: 15px; font-weight: 700; margin: 0;">📋 Character Information</h4><span id="icon_{name}" style="color: #1a9fff; font-weight: 700;">▼</span></div><div id="char_list_{name}" style="display: grid; grid-template-columns: 1fr 1fr; gap: 8px; margin-top: 10px; max-height: 180px; overflow-y: auto;">{''.join([f'''<div style="border-left: 4px solid #1a9fff; border-radius: 6px; padding: 8px; background: #f8fafc; cursor: pointer;" onclick="window.set_tts_val('{m[0]}')"><div style="font-weight: 700; color: #1e293b; font-size: 13px;">{m[0]}</div><div style="color: #94a3b8; font-size: 11px;">Character {i+1}</div></div>''' for i, m in enumerate(models)])}</div></div>""")
                                    # ALGORITMA OTOMATIS RMVPE
                                    f0method0 = gr.Radio(label="Algoritma Pitch", choices=f0method_mode, value="rmvpe" if "rmvpe" in f0method_mode else "pm")
                                    # RETRIEVAL OTOMATIS 0.75
                                    index_rate1 = gr.Slider(minimum=0, maximum=1, label="Rasio Retrieval", value=0.75)
                                    # MEDIAN FILTERING OTOMATIS 7
                                    filter_radius0 = gr.Slider(minimum=0, maximum=7, label="Median Filtering", value=7, step=1)
                                
                                with gr.Column():
                                    # RESAMPLE RATE OTOMATIS 0
                                    resample_sr0 = gr.Slider(minimum=0, maximum=48000, label="Resample Rate", value=0, step=1)
                                    # VOLUME ENVELOPE OTOMATIS 0.76
                                    rms_mix_rate0 = gr.Slider(minimum=0, maximum=1, label="Volume Envelope", value=0.76)
                                    # PROTEKSI SUARA OTOMATIS 0.33
                                    protect0 = gr.Slider(minimum=0, maximum=0.5, label="Proteksi Suara", value=0.33, step=0.01)
                                    
                                    # Notes Utama
                                    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>""")
                                    
                                    # Bagian 📑 DI SARANKAN 📑 (VERSI LENGKAP)
                                    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: #1a9fff; font-weight: 700; font-size: 13px;">🖥️ LOG SISTEM 🖥️</span></div>""")
                                    vc_log = gr.Textbox(label="", placeholder="📑 menunggu dibuatkan 📑", interactive=False)
                                    vc_output = gr.Audio(label="Audio Hasil", interactive=False)
                                    vc_convert = gr.Button("🎐 GENERATED VOICE 🎐", variant="primary")

                            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 #e0e6ed; border-radius: 12px; padding: 20px; background-color: white; text-align: center;"><h3 style="color: #1e293b; font-size: 18px; margin: 0; font-weight: 700; letter-spacing: 0.5px;">CREATED BY PLANA-CHAN</h3><p style="color: #94a3b8; font-size: 14px; margin-top: 4px; font-weight: 500;">Blue Archive TTS v4.0 • Powered by VITS</p></div><script>window.set_tts_val = function(val) { var textboxes = document.querySelectorAll('textarea[label="🏷️ MASUK TEXT SINI"]'); for (var t of textboxes) { t.value = val; t.dispatchEvent(new Event('input', { bubbles: true })); } }</script>""")
    
    app.queue(max_size=20).launch(share=False, server_name="0.0.0.0", server_port=7860)