Update app.py

app.py

@@ -1,101 +1,3 @@
-import os
-import re
-import time
-import sys
-import subprocess
-import scipy.io.wavfile as wavfile
-import torch
-import torchaudio
-import gradio as gr
-from TTS.api import TTS
-from TTS.tts.configs.xtts_config import XttsConfig
-from TTS.tts.models.xtts import Xtts
-from TTS.utils.generic_utils import get_user_data_dir
-from huggingface_hub import hf_hub_download
-
-# Configuración inicial
-os.environ["COQUI_TOS_AGREED"] = "1"
-
-def check_and_install(package):
-    try:
-        __import__(package)
-    except ImportError:
-        print(f"{package} no está instalado. Instalando...")
-        subprocess.check_call([sys.executable, "-m", "pip", "install", package])
-
-print("Descargando y configurando el modelo...")
-repo_id = "Blakus/Pedro_Lab_XTTS"
-local_dir = os.path.join(get_user_data_dir("tts"), "tts_models--multilingual--multi-dataset--xtts_v2")
-os.makedirs(local_dir, exist_ok=True)
-files_to_download = ["config.json", "model.pth", "vocab.json"]
-
-for file_name in files_to_download:
-    print(f"Descargando {file_name} de {repo_id}")
-    hf_hub_download(repo_id=repo_id, filename=file_name, local_dir=local_dir)
-
-config_path = os.path.join(local_dir, "config.json")
-checkpoint_path = os.path.join(local_dir, "model.pth")
-vocab_path = os.path.join(local_dir, "vocab.json")
-
-config = XttsConfig()
-config.load_json(config_path)
-
-model = Xtts.init_from_config(config)
-model.load_checkpoint(config, checkpoint_path=checkpoint_path, vocab_path=vocab_path, eval=True, use_deepspeed=True)
-
-model.cuda()
-
-print("Modelo cargado en GPU")
-
-def predict(prompt, language, reference_audio):
-    try:
-        if len(prompt) < 2 or len(prompt) > 600:
-            return None, "El texto debe tener entre 2 y 600 caracteres."
-
-        # Obtener los parámetros de la configuración JSON
-        temperature = config.model_args.get("temperature", 0.85)
-        length_penalty = config.model_args.get("length_penalty", 1.0)
-        repetition_penalty = config.model_args.get("repetition_penalty", 2.0)
-        top_k = config.model_args.get("top_k", 50)
-        top_p = config.model_args.get("top_p", 0.85)
-
-        gpt_cond_latent, speaker_embedding = model.get_conditioning_latents(
-            audio_path=reference_audio
-        )
-
-        start_time = time.time()
-
-        out = model.inference(
-            prompt,
-            language,
-            gpt_cond_latent,
-            speaker_embedding,
-            temperature=temperature,
-            length_penalty=length_penalty,
-            repetition_penalty=repetition_penalty,
-            top_k=top_k,
-            top_p=top_p
-        )
-
-        inference_time = time.time() - start_time
-        
-        output_path = "pedro_labattaglia_TTS.wav"
-        # Guardar el audio directamente desde el output del modelo
-        import scipy.io.wavfile as wavfile
-        wavfile.write(output_path, config.audio["output_sample_rate"], out["wav"])
-
-        audio_length = len(out["wav"]) / config.audio["output_sample_rate"]  # duración del audio en segundos
-        real_time_factor = inference_time / audio_length
-
-        metrics_text = f"Tiempo de generación: {inference_time:.2f} segundos\n"
-        metrics_text += f"Factor de tiempo real: {real_time_factor:.2f}"
-
-        return output_path, metrics_text
-
-    except Exception as e:
-        print(f"Error detallado: {str(e)}")
-        return None, f"Error: {str(e)}"
-
 # Configuración de la interfaz de Gradio
 supported_languages = ["es", "en"]
 reference_audios = [