app.py

import gradio as gr
import os
import subprocess
from gtts import gTTS
from pydub import AudioSegment
from transformers import GPT2LMHeadModel, GPT2Tokenizer, pipeline

model = GPT2LMHeadModel.from_pretrained("salomonsky/deepSP")
os.environ["TOKENIZERS_PARALLELISM"] = "true"
generator = pipeline('text-generation', model="salomonsky/deepSP")
tokenizer = GPT2Tokenizer.from_pretrained('salomonsky/deepSP')

def generate_output(text):
    prompt = ""
    input_tokens = tokenizer.encode(text, add_special_tokens=False)
    input_text = tokenizer.decode(input_tokens)
    gpt2_output = generator(input_text, max_length=20, do_sample=True, temperature=0.9)
    if len(gpt2_output) == 0 or 'generated_text' not in gpt2_output[0]:
        return None, "No se pudo generar el texto."

    generated_text = gpt2_output[0]['generated_text']
    generated_text = generated_text.replace(input_text, "").strip()

    try:
        tts = gTTS(generated_text, lang='es')
        temp_audio_path = "temp_audio.mp3"
        tts.save(temp_audio_path)
        audio_path = "audio.wav"
        audio = AudioSegment.from_mp3(temp_audio_path)
        audio.export(audio_path, format="wav")
        print("Archivo de audio generado:", audio_path)
    except Exception as e:
        return None, f"No se pudo generar el audio: {str(e)}"

    face_image_path = "face.jpg"
    if not os.path.isfile(face_image_path):
        return None, "No se encontró el archivo de imagen de cara."

    print("Archivo de imagen de cara:", face_image_path)

    command = f"python3 inference.py --checkpoint_path checkpoints/wav2lip_gan.pth --face {face_image_path} --audio {audio_path} --outfile video.mp4 --nosmooth"
    process = subprocess.run(command, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
    if process.returncode != 0:
        error_message = process.stderr.decode()
        return None, f"No se pudo generar el video: {error_message}"

    output_video_path = "video.mp4"
    print("Archivo de video generado:", output_video_path)

    os.remove(temp_audio_path)

    if os.path.isfile(output_video_path):
        return output_video_path, None

    return None, "No se pudo generar el video."

def error_message_fn(error_message):
    if error_message is not None:
        return gr.outputs.Textbox(text=error_message, placeholder="Error")
    else:
        return None

iface = gr.Interface(
    fn=generate_output,
    inputs=gr.inputs.Textbox(lines=1, placeholder='Escribe tu nombre para presentarte con Andrea...'),
    outputs=[
        gr.outputs.Video(label="Respuesta de Andrea (un minuto aproximadamente)"),
        gr.outputs.Textbox(label="Mensaje de error", type="text")
    ],
    title="Andrea - Humanoid Chatbot IA 2023(c)",
    error="No se pudo generar la salida.",
    error_message=error_message_fn
)

iface.launch()