app.py

import gradio as gr
# import os, subprocess, torchaudio
# import torch
from PIL import Image
from gtts import gTTS
import tempfile
from pydub import AudioSegment
from pydub.generators import Sine
# from fairseq.checkpoint_utils import load_model_ensemble_and_task_from_hf_hub
# from fairseq.models.text_to_speech.hub_interface import TTSHubInterface
import soundfile

import dlib
import cv2
import imageio
import os
import gradio as gr
import os, subprocess, torchaudio
from PIL import Image
import ffmpeg



block = gr.Blocks()

def merge_frames():
  path = '/content/video_results/restored_imgs'
  image_folder = os.fsencode(path)
  print(image_folder)
  filenames = []

  for file in os.listdir(image_folder):
      filename = os.fsdecode(file)
      if filename.endswith( ('.jpg', '.png', '.gif') ):
          filenames.append(filename)

  filenames.sort() # this iteration technique has no built in order, so sort the frames
  images = list(map(lambda filename: imageio.imread("/content/video_results/restored_imgs/"+filename), filenames))

  
  imageio.mimsave('/content/video_output.mp4', images, fps=25.0) # modify the frame duration as needed


block = gr.Blocks()



def audio_video():
  
  input_video = ffmpeg.input('/content/video_output.mp4')

  input_audio = ffmpeg.input('/content/audio.wav')

  ffmpeg.concat(input_video, input_audio, v=1, a=1).output('final_output.mp4').run()


def compute_aspect_preserved_bbox(bbox, increase_area, h, w):
    left, top, right, bot = bbox
    width = right - left
    height = bot - top

    width_increase = max(increase_area, ((1 + 2 * increase_area) * height - width) / (2 * width))
    height_increase = max(increase_area, ((1 + 2 * increase_area) * width - height) / (2 * height))

    left_t = int(left - width_increase * width)
    top_t = int(top - height_increase * height)
    right_t = int(right + width_increase * width)
    bot_t = int(bot + height_increase * height)

    left_oob = -min(0, left_t)
    right_oob = right - min(right_t, w)
    top_oob = -min(0, top_t)
    bot_oob = bot - min(bot_t, h)

    if max(left_oob, right_oob, top_oob, bot_oob) > 0:
        max_w = max(left_oob, right_oob)
        max_h = max(top_oob, bot_oob)
        if max_w > max_h:
            return left_t + max_w, top_t + max_w, right_t - max_w, bot_t - max_w
        else:
            return left_t + max_h, top_t + max_h, right_t - max_h, bot_t - max_h

    else:
        return (left_t, top_t, right_t, bot_t)

def crop_src_image(src_img, detector=None):
    if  detector is None:
        detector = dlib.get_frontal_face_detector()
    save_img='/content/image_pre.png'
    img = cv2.imread(src_img)
    faces = detector(img, 0)
    h, width, _ = img.shape
    if len(faces) > 0:
        bbox = [faces[0].left(), faces[0].top(),faces[0].right(), faces[0].bottom()]
        l = bbox[3]-bbox[1]
        bbox[1]= bbox[1]-l*0.1
        bbox[3]= bbox[3]-l*0.1
        bbox[1] = max(0,bbox[1])
        bbox[3] = min(h,bbox[3])
        bbox = compute_aspect_preserved_bbox(tuple(bbox), 0.5, img.shape[0], img.shape[1])
        img = img[bbox[1] :bbox[3] , bbox[0]:bbox[2]]
        img = cv2.resize(img, (256, 256))
        cv2.imwrite(save_img,img)
    else:
        img = cv2.resize(img,(256,256))
        cv2.imwrite(save_img, img)



def pad_image(image):
    w, h = image.size
    if w == h:
        return image
    elif w > h:
        new_image = Image.new(image.mode, (w, w), (0, 0, 0))
        new_image.paste(image, (0, (w - h) // 2))
        return new_image
    else:
        new_image = Image.new(image.mode, (h, h), (0, 0, 0))
        new_image.paste(image, ((h - w) // 2, 0))
        return new_image

def calculate(image_in, audio_in):
    waveform, sample_rate = torchaudio.load(audio_in)
    torchaudio.save("/content/audio.wav", waveform, sample_rate, encoding="PCM_S", bits_per_sample=16)
    image = Image.open(image_in)
    image = pad_image(image)
    image.save("image.png")

    pocketsphinx_run = subprocess.run(['pocketsphinx', '-phone_align', 'yes', 'single', '/content/audio.wav'], check=True, capture_output=True)
    jq_run = subprocess.run(['jq', '[.w[]|{word: (.t | ascii_upcase | sub("<S>"; "sil") | sub("<SIL>"; "sil") | sub("\\\(2\\\)"; "") | sub("\\\(3\\\)"; "") | sub("\\\(4\\\)"; "") | sub("\\\[SPEECH\\\]"; "SIL") | sub("\\\[NOISE\\\]"; "SIL")), phones: [.w[]|{ph: .t | sub("\\\+SPN\\\+"; "SIL") | sub("\\\+NSN\\\+"; "SIL"), bg: (.b*100)|floor, ed: (.b*100+.d*100)|floor}]}]'], input=pocketsphinx_run.stdout, capture_output=True)
    with open("test.json", "w") as f:
        f.write(jq_run.stdout.decode('utf-8').strip())

    os.system(f"cd /content/one-shot-talking-face && python3 -B test_script.py --img_path /content/results/restored_imgs/image_pre.png --audio_path /content/audio.wav --phoneme_path /content/test.json --save_dir /content/train")
    return "/content/train/image_audio.mp4"


def one_shot_talking(image_in,audio_in):

  #Pre-processing of image
  crop_src_image(image_in)

  #Improve quality of input image
  os.system(f"python /content/GFPGAN/inference_gfpgan.py --upscale 2 -i /content/image_pre.png -o /content/results  --bg_upsampler realesrgan")

  image_in_one_shot='/content/results/restored_imgs/image_pre.png'

  #One Shot Talking Face algorithm
  calculate(image_in_one_shot,audio_in)

  #Video Quality Improvement

  #1. Extract the frames from the video file using PyVideoFramesExtractor
  os.system(f"python /content/PyVideoFramesExtractor/extract.py --video=/content/train/image_pre_audio.mp4")

  #2. Improve image quality using GFPGAN on each frames
  os.system(f"python /content/GFPGAN/inference_gfpgan.py --upscale 2 -i /content/extracted_frames/image_pre_audio_frames -o /content/video_results  --bg_upsampler realesrgan")

  #3. Merge all the frames to a one video using imageio
  merge_frames()

  audio_video()
  return "Sucessufull"


def one_shot(image,input_text,gender): 
   if gender == 'Female' or gender == 'female':
     tts = gTTS(input_text)
     with tempfile.NamedTemporaryFile(suffix='.mp3', delete=False) as f:
          tts.write_to_fp(f)
          f.seek(0)
          sound = AudioSegment.from_file(f.name, format="mp3")
          sound.export("/content/audio.wav", format="wav")
     one_shot_talking(image,'audio.wav') 
       
   elif gender == 'Male' or gender == 'male':
      print(gender)
      models, cfg, task = load_model_ensemble_and_task_from_hf_hub(
          "Voicemod/fastspeech2-en-male1",
          arg_overrides={"vocoder": "hifigan", "fp16": False}
      )

      model = models[0].cuda()
      TTSHubInterface.update_cfg_with_data_cfg(cfg, task.data_cfg)
      generator = task.build_generator([model], cfg)
      # next(model.parameters()).device

      sample = TTSHubInterface.get_model_input(task, input_text)
      sample["net_input"]["src_tokens"] = sample["net_input"]["src_tokens"].cuda()
      sample["net_input"]["src_lengths"] = sample["net_input"]["src_lengths"].cuda()
      sample["speaker"] = sample["speaker"].cuda()

      wav, rate = TTSHubInterface.get_prediction(task, model, generator, sample)
      # soundfile.write("/content/audio_before.wav", wav, rate)
      soundfile.write("/content/audio_before.wav", wav.cpu().clone().numpy(), rate)
      cmd='ffmpeg -i /content/audio_before.wav -filter:a "atempo=0.7" -vn /content/audio.wav'
      os.system(cmd)
      one_shot_talking(image,'audio.wav')
    



input_value = "Hello How are you?"

image = gr.Image(show_label=True, type="filepath",label="Input Image")
input_text=gr.Textbox(lines=3, value=input_value, label="Input Text")
gender = gr.Radio(["Female","Male"],value="Female",label="Gender")
output = gr.Video(show_label=True,label="Output")

demo = gr.Interface(
    one_shot,
    [image,input_text,gender],
    [output],
    title="One Shot Talking Face from Text",
)
demo.launch(enable_queue = False)