app.py

import gradio as gr
import os
import re
import numpy as np
from TTS.utils.synthesizer import Synthesizer
from huggingface_hub import hf_hub_download

# 1. Maxfiy kalitni olish
hf_token = os.environ.get("HF_TOKEN")

# 2. Modelni maxfiy ombordan yuklab olish
repo_id = "UpCoder/behruz-vits-v3-private"

try:
    print("Model fayllari yuklanmoqda...")
    model_path = hf_hub_download(repo_id=repo_id, filename="checkpoint_43000.pth", token=hf_token)
    config_path = hf_hub_download(repo_id=repo_id, filename="config.json", token=hf_token)
except Exception as e:
    print(f"Fayllarni yuklashda xatolik: {e}")

# 3. Sun'iy intellektni ishga tushirish
print("Sun'iy intellekt ishga tushmoqda...")
synthesizer = Synthesizer(
    tts_checkpoint=model_path,
    tts_config_path=config_path,
    use_cuda=False 
)

# VITS uchun standart chastota (22050 Hz)
SAMPLE_RATE = 22050 

def split_into_sentences(text):
    sentences = re.split(r'(?<=[.!?]) +', text.strip())
    return [s for s in sentences if s.strip()]

def synthesize_full_audio(text):
    if not text.strip():
        return None
        
    # Xavfsizlik: Server qotib qolmasligi uchun matnni 2000 belgi bilan cheklaymiz
    if len(text) > 2000:
        text = text[:2000]
        
    sentences = split_into_sentences(text)
    all_wavs = []
    
    # Jumlalar orasida tabiiy nafas olish uchun 0.25 soniyalik sukut
    silence = np.zeros(int(SAMPLE_RATE * 0.25))
    
    for i, sentence in enumerate(sentences):
        try:
            wav = synthesizer.tts(sentence)
            all_wavs.append(np.array(wav))
            
            # Oxirgi jumladan tashqari hammadan keyin sukut qo'shamiz
            if i < len(sentences) - 1:
                all_wavs.append(silence)
                
        except Exception as e:
            print(f"Jumlani o'qishda xatolik: {sentence}. Xato: {e}")
            continue
            
    if not all_wavs:
        return None
        
    # Barcha audio parchalarni bitta butun faylga birlashtirish
    final_wav = np.concatenate(all_wavs)
    final_wav_int16 = (final_wav * 32767).astype(np.int16)
    
    return (SAMPLE_RATE, final_wav_int16)

# 4. Professional va O'zbekcha Interfeys (UI) yaratish
with gr.Blocks(theme=gr.themes.Soft(primary_hue="emerald", secondary_hue="teal")) as iface:
    gr.Markdown(
        """
        <div style="text-align: center;">
            <h1>🎙️ Behruzning Raqamli Ovozli Kloni (V3)</h1>
            <p><strong>Mening sun'iy intellekt ovoz generatorimga xush kelibsiz!</strong> Ushbu model o'zimning haqiqiy ovozim asosida neyrotarmoqlar yordamida o'qitildi.</p>
        </div>
        
        💡 **Foydali maslahat:** Katta matnlarni (masalan, butun bir xatboshini) bemalol kiritishingiz mumkin! Dastur uni avtomat ravishda jumlalarga bo'lib, xatosiz o'qib beradi va bitta tayyor audio fayl qilib taqdim etadi.
        """
    )
    
    with gr.Row():
        with gr.Column(scale=2):
            text_input = gr.Textbox(
                label="O'zbekcha matnni bu yerga kiriting (Maksimum 2000 belgi)", 
                lines=6, 
                placeholder="Salom! Bugun havo juda ajoyib, shunday emasmi? Men internetda yashaydigan raqamli sun'iy intellektman..."
            )
            generate_btn = gr.Button("🚀 Ovozga Aylantirish", variant="primary")
            
        with gr.Column(scale=1):
            audio_output = gr.Audio(label="🎧 Tayyor Audio Fayl")
            
    gr.Examples(
        examples=[
            "Salom, men Behruzning raqamli egizagiman va men endi internetda yashayman!",
            "Axborot texnologiyalari sohasida qanday yangiliklar bor, kuzatib boryapsizmi?",
            "Voh, bu natijani umuman kutmagan edim! Qoyilmaqom ish bo'libdi."
        ],
        inputs=text_input,
        label="Namuna jumlalar (birini tanlang)"
    )

    generate_btn.click(fn=synthesize_full_audio, inputs=text_input, outputs=audio_output)

iface.launch()