Update app.py

app.py

@@ -1,71 +1,5 @@
-import sys
-import io, os, stat
-import subprocess
-import random
-from zipfile import ZipFile
-import uuid
 import time
-import torch
-import torchaudio
 
-# Mantenemos la descarga de MeCab
-os.system('python -m unidic download')
-
-# Mantenemos el acuerdo de CPML
-os.environ["COQUI_TOS_AGREED"] = "1"
-
-import langid
-import base64
-import csv
-from io import StringIO
-import datetime
-import re
-
-import gradio as gr
-from scipy.io.wavfile import write
-from pydub import AudioSegment
-
-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
-
-HF_TOKEN = os.environ.get("HF_TOKEN")
-
-from huggingface_hub import hf_hub_download
-import os
-from TTS.utils.manage import get_user_data_dir
-
-# Mantenemos la autenticación y descarga del 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"Downloading {file_name} from {repo_id}")
-    local_file_path = os.path.join(local_dir, file_name)
-    hf_hub_download(repo_id=repo_id, filename=file_name, local_dir=local_dir)
-
-# Cargamos configuración y modelo
-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=False)
-
-print("Modelo cargado en CPU")
-
-# Mantenemos variables globales y funciones auxiliares
-DEVICE_ASSERT_DETECTED = 0
-DEVICE_ASSERT_PROMPT = None
-DEVICE_ASSERT_LANG = None
-supported_languages = config.languages
-
-# Función de inferencia usando parámetros predeterminados del archivo de configuración
 def predict(prompt, language, audio_file_pth, mic_file_path, use_mic):
     try:
         if use_mic:
@@ -90,6 +24,8 @@ def predict(prompt, language, audio_file_pth, mic_file_path, use_mic):
             max_ref_length=max_ref_length
         )
 
+        # Medimos el tiempo de inferencia manualmente
+        start_time = time.time()
         out = model.inference(
             prompt,
             language,
@@ -98,43 +34,19 @@ def predict(prompt, language, audio_file_pth, mic_file_path, use_mic):
             temperature=temperature,
             repetition_penalty=repetition_penalty,
         )
+        inference_time = time.time() - start_time
         
         torchaudio.save("output.wav", torch.tensor(out["wav"]).unsqueeze(0), 24000)
 
-        metrics_text = f"Tiempo de generación: {out['inference_time']:.2f} segundos\n"
-        metrics_text += f"Factor de tiempo real: {out['inference_time'] / (len(out['wav']) / 24000):.2f}"
+        # Calculamos las métricas usando el tiempo medido manualmente
+        audio_length = len(out["wav"]) / 24000  # 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 gr.make_waveform("output.wav"), "output.wav", metrics_text
 
     except Exception as e:
         print(f"Error detallado: {str(e)}")
-        return None, None, f"Error: {str(e)}"
-
-# Interfaz de Gradio actualizada sin sliders
-with gr.Blocks(theme=gr.themes.Base()) as demo:
-    gr.Markdown("# Sintetizador de Voz XTTS")
-    
-    with gr.Row():
-        with gr.Column():
-            input_text = gr.Textbox(label="Texto a sintetizar", placeholder="Escribe aquí el texto que quieres convertir a voz...")
-            language = gr.Dropdown(label="Idioma", choices=supported_languages, value="es")
-            audio_file = gr.Audio(label="Audio de referencia", type="filepath")
-            use_mic = gr.Checkbox(label="Usar micrófono")
-            mic_file = gr.Audio(label="Grabar con micrófono", source="microphone", type="filepath", visible=False)
-            
-            use_mic.change(fn=lambda x: gr.update(visible=x), inputs=[use_mic], outputs=[mic_file])
-            
-            generate_button = gr.Button("Generar voz")
-        
-        with gr.Column():
-            output_audio = gr.Audio(label="Audio generado")
-            waveform = gr.Image(label="Forma de onda")
-            metrics = gr.Textbox(label="Métricas")
-    
-    generate_button.click(
-        predict,
-        inputs=[input_text, language, audio_file, mic_file, use_mic],
-        outputs=[waveform, output_audio, metrics]
-    )
-
-demo.launch(debug=True)
+        return None, None, f"Error: {str(e)}"