app.py

import streamlit as st
import os
import sys
import torch
import pickle
import numpy
import librosa
import subprocess
from avatar import Avatar

def run_pickleface():
    try:
        result = subprocess.run(
            ['python', 'pickleface.py'],
            check=True,
            capture_output=True,
            text=True
        )
        print(result.stdout)
        if result.returncode != 0:
            st.error(f"Error creating face detection results: {result.stderr}")
            return False
        return True
    except subprocess.CalledProcessError as e:
        st.error(f"Critical error running pickleface.py: {e.stderr}")
        return False

def initialize_face_detection_results():
    # Kiểm tra xem tất cả file pkl đã tồn tại chưa
    missing_files = [opt for opt in options if not os.path.exists(f'ref_videos/{opt}_face_det_result.pkl')]
    
    if missing_files:
        current_status_placeholder.write("Creating face detection results...")
        if not run_pickleface():
            st.error("Failed to create face detection results")
            st.stop()
        current_status_placeholder.write("Face detection results created successfully!")

# Cấu hình ban đầu
options = ['Aude', 'Kyla', 'Liv', 'MC6']
images = ['ref_videos/Aude.png', 'ref_videos/Kyla.png', 'ref_videos/Liv.png', 'ref_videos/MC6.png']

# Thêm đường dẫn đến thư mục Wav2Lip
wav2lip_path = os.path.join(os.path.dirname(__file__), "Wav2Lip")
if wav2lip_path not in sys.path:
    sys.path.insert(0, wav2lip_path)

# Giao diện
big_text = """
    <div style='text-align: center;'>
        <h1 style='font-size: 30x;'>Text to Speech Synchronized Video</h1>
    </div>
"""
st.markdown(big_text, unsafe_allow_html=True)
current_status_placeholder = st.empty()
init_progress_bar = st.progress(0)

# Khởi tạo session state
if 'is_initialized' not in st.session_state:
    initialize_face_detection_results()
    
    # Khởi tạo Avatar
    st.session_state.avatar = Avatar()
    st.session_state.avatar.export_video = False
    
    # Load model
    current_status_placeholder.write("Loading model...")
    st.session_state.avatar.load_model("checkpoint/wav2lip_gan.pth")
    current_status_placeholder.write("Model loaded successfully")
    
    # Cấu hình thiết bị
    st.session_state.avatar.device = 'cuda' if torch.cuda.is_available() else 'cpu'
    print(f"Using device: {st.session_state.avatar.device}")
    
    # Cấu hình đường dẫn
    st.session_state.avatar.output_audio_path = "audio/"
    st.session_state.avatar.output_audio_filename = "result.wav"
    st.session_state.avatar.temp_lip_video_no_voice_path = "temp/"
    st.session_state.avatar.temp_lip_video_no_voice_filename = "result.avi"
    st.session_state.avatar.output_video_path = "results/"
    st.session_state.avatar.output_video_name = "result_voice.mp4"
    
    # Khởi tạo video mặc định
    st.session_state.selected_option = "Liv"
    st.session_state.avatar.ref_video_path_and_filename = f"ref_videos/{st.session_state.selected_option}.mp4"
    
    # Xử lý video và face detection
    st.session_state.avatar.get_video_full_frames(st.session_state.avatar.ref_video_path_and_filename)
    st.session_state.avatar.face_detect_batch_size = 16
    
    # Load face detection results cho tất cả options
    st.session_state.face_det_results_dict = {}
    for option in options:
        with open(f'ref_videos/{option}_face_det_result.pkl', 'rb') as file:
            st.session_state.face_det_results_dict[option] = pickle.load(file)
    
    st.session_state.avatar.face_detect_img_results = st.session_state.face_det_results_dict[st.session_state.selected_option]
    st.session_state.avatar.face_det_results_path_and_name = f'ref_videos/{st.session_state.selected_option}_face_det_result.pkl'
    
    # Xử lý text to speech
    input_text = "Hi How are you?"
    st.session_state.avatar.text_to_lip_video(input_text, init_progress_bar)
    current_status_placeholder.write("Face detection results loaded")
    
    st.session_state['is_initialized'] = True

# Giao diện lựa chọn video
selected_option = st.radio("Choose an option:", options, index=options.index(st.session_state.selected_option))
img_col1, img_col2 = st.columns([1,1])

with img_col1:
    st.image(images[options.index(selected_option)])

# Xử lý khi thay đổi lựa chọn video
if st.session_state.selected_option != selected_option:
    print("The selected option has changed!")
    st.session_state.selected_option = selected_option
    st.session_state.avatar.ref_video_path_and_filename = f"ref_videos/{st.session_state.selected_option}.mp4"
    
    st.session_state.avatar.get_video_full_frames(st.session_state.avatar.ref_video_path_and_filename)
    st.session_state.avatar.face_detect_img_results = st.session_state.face_det_results_dict[st.session_state.selected_option]