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 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")
          return sound.export("/content/audio.wav", format="wav")
          exit()
     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')

       


def generate_ocr(method,image,gender):
    return "Hello"
    
def run():
  with block:
  
    with gr.Group():
      with gr.Box():
        with gr.Row().style(equal_height=True):
          image_in = gr.Image(show_label=False, type="filepath")
          # audio_in = gr.Audio(show_label=False, type='filepath')
          input_text=gr.Textbox(lines=3, value="Hello How are you?", label="Input Text")
          gender = gr.Radio(["Female","Male"],value="Female",label="Gender")
          video_out = gr.Audio(label="output")
          # video_out = gr.Video(show_label=False)
        with gr.Row().style(equal_height=True):
          btn = gr.Button("Generate")          

    btn.click(one_shot, inputs=[image_in, input_text,gender], outputs=[video_out])
    # block.queue()
    block.launch(server_name="0.0.0.0", server_port=7860)

if __name__ == "__main__":
    run()


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

# demo = gr.Interface(generate_ocr,[image,input_text,gender],[output],title="One Shot Talking Face from Text",)
# demo.launch(server_name="0.0.0.0", server_port=7860)