Update app.py

app.py

@@ -49,4 +49,142 @@ def setup_mecab_and_unidic():
 print("Configurando MeCab y UniDic...")
 setup_mecab_and_unidic()
 
-# El resto del código permanece igual...
+# Descargar y configurar el modelo
+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=False)
+
+print("Modelo cargado en CPU")
+
+# Funciones auxiliares
+def split_text(text):
+    return re.split(r'(?<=[.!?])\s+', text)
+
+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."
+
+        sentences = split_text(prompt)
+
+        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=reference_audio,
+            gpt_cond_len=gpt_cond_len,
+            gpt_cond_chunk_len=gpt_cond_chunk_len,
+            max_ref_length=max_ref_length
+        )
+
+        start_time = time.time()
+        combined_audio = AudioSegment.empty()
+
+        for sentence in sentences:
+            out = model.inference(
+                sentence,
+                language,
+                gpt_cond_latent,
+                speaker_embedding,
+                temperature=temperature,
+                repetition_penalty=repetition_penalty,
+            )
+            audio_segment = AudioSegment(
+                out["wav"].tobytes(),
+                frame_rate=24000,
+                sample_width=2,
+                channels=1
+            )
+            combined_audio += audio_segment
+            combined_audio += AudioSegment.silent(duration=500)  # 0.5 segundos de silencio
+
+        inference_time = time.time() - start_time
+        
+        output_path = "output.wav"
+        combined_audio.export(output_path, format="wav")
+
+        audio_length = len(combined_audio) / 1000  # 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 = [
+    "serio.wav",
+    "neutral.wav",
+    "alegre.wav",
+]
+
+theme = gr.themes.Soft(
+    primary_hue="blue",
+    secondary_hue="gray",
+).set(
+    body_background_fill='*neutral_100',
+    body_background_fill_dark='*neutral_900',
+)
+
+description = """
+# Sintetizador de voz de Pedro Labattaglia 🎙️
+
+Sintetizador de voz con la voz del locutor argentino Pedro Labattaglia. 
+
+## Cómo usarlo:
+- Elija el idioma (Español o Inglés)
+- Elija un audio de referencia de la lista 
+- Escriba el texto que desea sintetizar
+- Presione generar voz
+"""
+
+# Interfaz de Gradio
+with gr.Blocks(theme=theme) as demo:
+    gr.Markdown(description)
+
+    with gr.Row():
+        gr.Image("https://i1.sndcdn.com/artworks-000237574740-gwz61j-t500x500.jpg", label="", show_label=False, width=250, height=250)
+
+    with gr.Row():
+        with gr.Column(scale=2):
+            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):
+            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_selector, reference_audio],
+        outputs=[generated_audio, metrics_output]
+    )
+
+if __name__ == "__main__":
+    demo.launch()