app.py

import streamlit as st
import os
import sys
import torch
import pickle
import numpy
print(numpy.__version__)
import librosa
print(librosa.__version__)
import numpy
print(numpy.__version__)
path_to_add = os.path.join(os.path.dirname(__file__), "Wav2Lip")
if path_to_add not in sys.path:
    sys.path.insert(0, path_to_add)
from avatar import Avatar

options = ['Aude', 'Kyla', 'Liv', 'MC6']
images = ['ref_videos/Aude.png', 'ref_videos/Kyla.png', 'ref_videos/Liv.png', 'ref_videos/MC6.png']
big_text = """
    <div style='text-align: center;'>
        <h1 style='font-size: 30x;'>Text to Speech Synchronized Video</h1>
    </div>
    """
    # Display the styled text
st.markdown(big_text, unsafe_allow_html=True)
current_status_placeholder = st.empty()
init_progress_bar = st.progress(0)
if 'is_initialized' not in st.session_state:
    st.session_state.avatar = Avatar()
    st.session_state.avatar.export_video = False
    current_status_placeholder.write("load model")
    st.session_state.avatar.load_model("checkpoint/wav2lip_gan.pth")
    current_status_placeholder.write("load model finished")
    st.session_state.avatar.device = 'cuda' if torch.cuda.is_available() else 'cpu'
    print(st.session_state.avatar.device)
    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"
    st.session_state.selected_option = "Liv"
    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_batch_size = 16
    st.session_state.avatar.create_face_detection_results(st.session_state.avatar.video_full_frames, True)
    current_status_placeholder.write("load face detection result")
    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 = 'ref_videos/Liv_face_det_result.pkl'

    st.session_state.avatar.load_face_detection_results()
    def load_face_detection_results(self):
         with open(self.face_det_results_path_and_name, 'rb') as file:
           self.face_detect_img_results = pickle.load(file)
    input_text = "Hi How are you?"
    st.session_state.avatar.text_to_lip_video(input_text,init_progress_bar)
    current_status_placeholder.write("load face detection result done")

    st.session_state['is_initialized'] = True




# Create the radio button group
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)])


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]

from avatar import Avatar
# Create a text input box and store the input in a variable
user_input = st.text_input("Enter your text:")
inference_progress_bar = st.progress(0)
if user_input:
    st.session_state.avatar.dir_clean_up()
    # Display the entered text
    st.write("You entered:", user_input)
    st.session_state.avatar.export_video=True
    st.session_state.avatar.text_to_lip_video(user_input,inference_progress_bar)
    col1, col2, col3 = st.columns([1, 4, 1])

    # with col1:
    #     st.write("Column 1 content")

    with col2:
        st.video(st.session_state.avatar.output_video_path + st.session_state.avatar.output_video_name)
    # with col3:
    #     st.write("Column 3 content")