Update app.py

app.py

@@ -1,22 +1,8 @@
-import sys
-import io, os, stat
-import subprocess
-import random
-from zipfile import ZipFile
-import uuid
-import time
-import torch
-import torchaudio
-import langid
-import base64
-import csv
-from io import StringIO
-import datetime
+import os
 import re
-from scipy.io.wavfile import write
-from pydub import AudioSegment
-
+import time
 import gradio as gr
+from pydub import AudioSegment
 from TTS.api import TTS
 from TTS.tts.configs.xtts_config import XttsConfig
 from TTS.tts.models.xtts import Xtts
@@ -25,18 +11,18 @@ from huggingface_hub import hf_hub_download
 
 # Configuración inicial
 os.environ["COQUI_TOS_AGREED"] = "1"
+os.system('python -m unidic download')
 
-# Autenticación y descarga del modelo
+# Descargar y configurar 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"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)
 
-# Carga de 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")
@@ -49,37 +35,25 @@ model.load_checkpoint(config, checkpoint_path=checkpoint_path, vocab_path=vocab_
 
 print("Modelo cargado en CPU")
 
-# Variables globales
-supported_languages = [lang for lang in config.languages if lang != "ja"]
-reference_audios = [
-    "serio.wav",
-    "neutral.wav",
-    "alegre.wav",
-]
-
-# Función para dividir el texto en chunks
+# Funciones auxiliares
 def split_text(text):
-    sentences = re.split(r'(?<=[.!?])\s+', text)
-    return sentences
+    return re.split(r'(?<=[.!?])\s+', text)
 
-# Función de inferencia mejorada
-def predict(prompt, language, audio_file_pth, use_reference_audio):
+def predict(prompt, language, reference_audio):
     try:
-        if use_reference_audio:
-            speaker_wav = audio_file_pth
-        else:
-            speaker_wav = "neutral.wav"  # Audio por defecto si no se selecciona uno
+        if len(prompt) < 2 or len(prompt) > 600:
+            return None, "El texto debe tener entre 2 y 600 caracteres."
 
         sentences = split_text(prompt)
-        
-        temperature = getattr(config, "temperature", 0.75)
-        repetition_penalty = getattr(config, "repetition_penalty", 5.0)
-        gpt_cond_len = getattr(config, "gpt_cond_len", 30)
-        gpt_cond_chunk_len = getattr(config, "gpt_cond_chunk_len", 4)
-        max_ref_length = getattr(config, "max_ref_len", 60)
+
+        temperature = config.inference.get("temperature", 0.75)
+        repetition_penalty = config.inference.get("repetition_penalty", 5.0)
+        gpt_cond_len = config.inference.get("gpt_cond_len", 30)
+        gpt_cond_chunk_len = config.inference.get("gpt_cond_chunk_len", 4)
+        max_ref_length = config.inference.get("max_ref_length", 60)
 
         gpt_cond_latent, speaker_embedding = model.get_conditioning_latents(
-            audio_path=speaker_wav,
+            audio_path=reference_audio,
             gpt_cond_len=gpt_cond_len,
             gpt_cond_chunk_len=gpt_cond_chunk_len,
             max_ref_length=max_ref_length
@@ -87,7 +61,7 @@ def predict(prompt, language, audio_file_pth, use_reference_audio):
 
         start_time = time.time()
         combined_audio = AudioSegment.empty()
-        
+
         for sentence in sentences:
             out = model.inference(
                 sentence,
@@ -117,13 +91,20 @@ def predict(prompt, language, audio_file_pth, use_reference_audio):
         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_path), output_path, metrics_text
+        return output_path, metrics_text
 
     except Exception as e:
         print(f"Error detallado: {str(e)}")
-        return None, None, f"Error: {str(e)}"
+        return None, f"Error: {str(e)}"
+
+# Configuración de la interfaz de Gradio
+supported_languages = ["es", "en"]
+reference_audios = [
+    "serio.wav",
+    "neutral.wav",
+    "alegre.wav",
+]
 
-# Definir el tema personalizado
 theme = gr.themes.Soft(
     primary_hue="blue",
     secondary_hue="gray",
@@ -132,16 +113,15 @@ theme = gr.themes.Soft(
     body_background_fill_dark='*neutral_900',
 )
 
-# Descripción del proyecto
 description = """
 # Sintetizador de voz de Pedro Labattaglia 🎙️
 
 Sintetizador de voz con la voz del locutor argentino Pedro Labattaglia. 
 
 ## Cómo usarlo:
-- Elija el idioma
-- Elija un audio de referencia de la lista o cargue su propio audio
-- Escriba el texto a sintetizar
+- Elija el idioma (Español o Inglés)
+- Elija un audio de referencia de la lista 
+- Escriba el texto que desea sintetizar
 - Presione generar voz
 """
 
@@ -154,30 +134,20 @@ with gr.Blocks(theme=theme) as demo:
 
     with gr.Row():
         with gr.Column(scale=2):
-            language = gr.Dropdown(label="Idioma", choices=supported_languages, value="es")
-            use_reference_audio = gr.Checkbox(label="Usar audio de referencia")
-            reference_audio = gr.Dropdown(label="Audio de referencia predefinido", choices=reference_audios, visible=False)
-            audio_file = gr.Audio(label="O cargue su propio audio de referencia", type="filepath", visible=False)
-            
-            use_reference_audio.change(
-                fn=lambda x: [gr.update(visible=x), gr.update(visible=x)],
-                inputs=[use_reference_audio],
-                outputs=[reference_audio, audio_file]
-            )
-            
-            input_text = gr.Textbox(label="Texto a sintetizar", placeholder="Escribe aquí el texto que quieres convertir a voz...", lines=5)
+            language_selector = gr.Dropdown(label="Idioma", choices=supported_languages)
+            reference_audio = gr.Dropdown(label="Audio de referencia", choices=reference_audios)
+            input_text = gr.Textbox(label="Texto a sintetizar", placeholder="Escribe aquí el texto que quieres convertir a voz...")
             generate_button = gr.Button("Generar voz", variant="primary")
 
         with gr.Column(scale=1):
-            output_audio = gr.Audio(label="Audio generado")
-            waveform = gr.Image(label="Forma de onda")
-            metrics = gr.Textbox(label="Métricas")
-    
+            generated_audio = gr.Audio(label="Audio generado", interactive=False)
+            metrics_output = gr.Textbox(label="Métricas", value="Tiempo de generación: -- segundos\nFactor de tiempo real: --")
+
     generate_button.click(
         predict,
-        inputs=[input_text, language, audio_file, use_reference_audio],
-        outputs=[waveform, output_audio, metrics]
+        inputs=[input_text, language_selector, reference_audio],
+        outputs=[generated_audio, metrics_output]
     )
 
 if __name__ == "__main__":
-    demo.launch(debug=True)
+    demo.launch()