app.py

import json
import os
import re
import logging
import random
import shutil
import numpy as np
import torch
from torch import no_grad, LongTensor
import commons
import utils
import gradio as gr
from models import SynthesizerTrn
from text import text_to_sequence, _clean_text
from huggingface_hub import hf_hub_download

# --- 1. SETUP LOGGING ---
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)

# --- 2. CONFIGURATION & ASSETS ---
HF_TOKEN = os.getenv("HF_TOKEN")
REPO_ID = "Plana-Archive/Plana-TTS"
SUBFOLDER = "Prosekai-TTS/saved_model"
device = torch.device("cpu")

# --- 3. ROMAJI CONVERTER ---
try:
    import pykakasi
    kks = pykakasi.kakasi()
    def to_romaji(text):
        if not text or text == "None": return ""
        try:
            result = kks.convert(str(text))
            return "".join([item['hepburn'].capitalize() for item in result])
        except:
            return str(text)
except:
    def to_romaji(text): return str(text)

# --- 4. LOADING MODEL ---
logger.info("[*] Downloading Project Sekai model assets...")
config_path = hf_hub_download(repo_id=REPO_ID, filename="config.json", subfolder=SUBFOLDER, token=HF_TOKEN)
model_path = hf_hub_download(repo_id=REPO_ID, filename="model.pth", subfolder=SUBFOLDER, token=HF_TOKEN)
cover_path = hf_hub_download(repo_id=REPO_ID, filename="cover.png", subfolder=SUBFOLDER, token=HF_TOKEN)

hps = utils.get_hparams_from_file(config_path)
model = SynthesizerTrn(
    len(hps.symbols),
    hps.data.filter_length // 2 + 1,
    hps.train.segment_size // hps.data.hop_length,
    n_speakers=hps.data.n_speakers,
    **hps.model).to(device)

utils.load_checkpoint(model_path, model, None)
model.eval()

speaker_names = [name for name in hps.speakers if name != "None"]
display_names = [to_romaji(name) for name in speaker_names]
speaker_map = {romaji: original for romaji, original in zip(display_names, speaker_names)}

# --- 5. LOGIC FUNCTIONS ---
def get_text(text, hps, is_phoneme):
    text_norm = text_to_sequence(text, hps.symbols, [] if is_phoneme else hps.data.text_cleaners)
    if hps.data.add_blank:
        text_norm = commons.intersperse(text_norm, 0)
    return LongTensor(text_norm)

def tts_execute(text, speaker_romaji, speed, is_phoneme):
    if not speaker_romaji:
        return "❌ Pilih Karakter dulu!", None
    try:
        original_name = speaker_map[speaker_romaji]
        speaker_id = hps.speakers.index(original_name)
        stn_tst = get_text(text, hps, is_phoneme)
        with no_grad():
            x_tst = stn_tst.unsqueeze(0).to(device)
            x_tst_lengths = LongTensor([stn_tst.size(0)]).to(device)
            sid = LongTensor([speaker_id]).to(device)
            audio = model.infer(x_tst, x_tst_lengths, sid=sid, noise_scale=.667, 
                                noise_scale_w=0.8, length_scale=1.0 / speed)[0][0, 0].data.cpu().float().numpy()
        return "✅ Success!", (hps.data.sampling_rate, (audio * 32767).astype(np.int16))
    except Exception as e:
        return f"Error: {e}", None

def get_random_jp():
    return random.choice(["こんにちは！", "お元気ですか？", "ワンダショ、最高！", "練習、頑張ろうね。", "また明日も会えるかな？"])

def to_phoneme_fn(text):
    return _clean_text(text, hps.data.text_cleaners) if text != "" else ""

# --- 6. UI STYLE ---
css = """
:root { 
    --primary-600: #1299ff !important; 
    --accent-600: #1299ff !important; 
    --loader-color: #add8e6 !important; /* Warna Biru Muda untuk Loading */
}

/* Modifikasi Loading Screen Gradio */
.gradio-container .load-overlay {
    background: rgba(255, 255, 255, 0.8) !important;
}
.gradio-container .loader {
    border-top-color: #add8e6 !important;
}

.ba-header-container { border: 1.5px solid #e1e8f0; border-radius: 12px; padding: 20px 10px; margin-bottom: 12px; background: white; text-align: center; }
.ba-header-container h1 { color: #1299ff !important; font-weight: 700 !important; font-size: 36px !important; margin: 0; }
.status-container { border: 1.5px solid #e1e8f0; border-radius: 12px; padding: 15px 22px; margin-bottom: 20px; background: white; }
.status-title { color: #1299ff !important; font-weight: 800; font-size: 16px; margin-bottom: 8px; }
.text-green-bold { color: #28a745 !important; font-weight: 900 !important; }
.text-blue-status { color: #1299ff !important; }
.slim-card { max-width: 500px; margin: 0 auto; background: transparent; padding: 10px; }
.scroll-box { height: 220px; overflow-y: auto; border: 1px solid #f0f4f8; border-radius: 12px; padding: 10px; background: #fafbfc; margin-bottom: 10px; }
.char-btn { background: white !important; border: 1px solid #e2e8f0 !important; border-left: 5px solid #1299ff !important; text-align: left !important; padding: 8px !important; font-size: 12px !important; margin-bottom: 4px !important; width: 100%; color: #4a5568 !important; }
.warning-card { background: #fff9f0; border: 2px dashed #f5a623; border-radius: 10px; padding: 12px; margin-bottom: 15px; text-align: center; }
.jp-btn { background: #f8fafc !important; border: 1px solid #cbd5e1 !important; color: #475569 !important; font-weight: 700 !important; border-radius: 10px !important; margin-bottom: 10px; font-size: 12px !important; width: 100%; }
.gen-btn { background: #1299ff !important; color: white !important; font-weight: 700 !important; border-radius: 12px !important; height: 45px !important; width: 100%; border: none !important; cursor: pointer; }
.credit-footer { margin-top: 25px; padding: 15px; background: white; border-radius: 12px; text-align: center; border-bottom: 4px solid #1299ff; color: #94a3b8; font-weight: 700; font-size: 12px; letter-spacing: 2px; }
"""

# --- 7. GRADIO INTERFACE ---
with gr.Blocks(title="Project Sekai TTS", css=css) as app:
    with gr.Column(elem_classes="slim-card"):
        gr.HTML(f"""
            <div class="ba-header-container">
                <h1>Project Sekai</h1>
                <p>💫 VITS Emotional TTS 💫</p>
            </div>
            <div class="status-container">
                <div class="status-title">System Status</div>
                <div class="status-item"><span style="color:#4a5568">Model :</span> <span class="text-green-bold">&nbsp;ProSekai Pack ✅</span></div>
                <div class="status-item"><span style="color:#4a5568">Device :</span> <span class="text-blue-status">&nbsp;CPU Mode</span></div>
            </div>
        """)

        if os.path.exists(cover_path):
            gr.Image(cover_path, show_label=False, interactive=False, height=160)

        sel_name = gr.State("")
        char_display = gr.Markdown("📍 *Silakan pilih karakter...*")
        
        with gr.Column(elem_classes="scroll-box"):
            for name in display_names:
                btn = gr.Button(f"👤 {name}", elem_classes="char-btn")
                btn.click(fn=lambda n=name: (n, f"📍 Selected: **{n}**"), outputs=[sel_name, char_display])

        with gr.Column():
            gr.HTML("""
                <div class="warning-card">
                    <div style="color:#f5a623;font-weight:800;font-size:13px;">🔖 INSTRUKSI 🔖</div>
                    <div style="color:#855d1a;font-size:11px;font-weight:600;">
                        Pilih member Proseka, tulis teks Jepang, lalu klik Generate!
                    </div>
                </div>
            """)
            
            txt_in = gr.TextArea(label="Input Text", value="こんにちは。", lines=3)
            gr.Button("🎲 RANDOM JAPANESE TEXT 🎲", elem_classes="jp-btn").click(get_random_jp, outputs=[txt_in])
            speed = gr.Slider(0.5, 2.0, 1.0, step=0.1, label="Speed Audio")
            
            with gr.Accordion(label="Advanced Options", open=False):
                phoneme_input = gr.Checkbox(value=False, label="Phoneme input")
                to_phoneme_btn = gr.Button("Convert text to phoneme")
                
                phoneme_list = gr.Dataset(
                    label="Phoneme list", 
                    components=[txt_in], 
                    samples=[[x] for x in hps.symbols]
                )
                to_phoneme_btn.click(to_phoneme_fn, [txt_in], [txt_in])

            btn_gen = gr.Button("🎐 GENERATE VOICE 🎐", elem_classes="gen-btn")
            status_out = gr.Textbox(label="Status Message", interactive=False)
            aud_out = gr.Audio(label="Voice Output")

            btn_gen.click(
                fn=tts_execute, 
                inputs=[txt_in, sel_name, speed, phoneme_input], 
                outputs=[status_out, aud_out]
            )

        gr.HTML("""<div class="credit-footer">🥒 CREATED BY MUTSUMI 🥒</div>""")

if __name__ == "__main__":
    app.launch(
        server_name="0.0.0.0", 
        server_port=7860,
        ssr_mode=False
    )