app.py

import spaces
import gradio as gr
import torch
from TTS.api import TTS
import os
import tempfile
import torchaudio
from huggingface_hub import hf_hub_download
from TTS.tts.configs.xtts_config import XttsConfig
from TTS.tts.models.xtts import Xtts

# Aceptar los términos de COQUI
os.environ["COQUI_TOS_AGREED"] = "1"

# Definir el dispositivo como CPU
device = "cpu"

# Descargar archivos desde HuggingFace
model_path = hf_hub_download(repo_id="RedSparkie/danielmula", filename="model.pth")
config_path = hf_hub_download(repo_id="RedSparkie/danielmula", filename="config.json")
vocab_path = hf_hub_download(repo_id="RedSparkie/danielmula", filename="vocab.json")

# Función para limpiar la caché de GPU (por si en el futuro se usa GPU)
def clear_gpu_cache():
    if torch.cuda.is_available():
        torch.cuda.empty_cache()

# Cargar el modelo XTTS
XTTS_MODEL = None
def load_model(xtts_checkpoint, xtts_config, xtts_vocab):
    global XTTS_MODEL
    clear_gpu_cache()
    if not xtts_checkpoint or not xtts_config or not xtts_vocab:
        return "You need to run the previous steps or manually set the `XTTS checkpoint path`, `XTTS config path`, and `XTTS vocab path` fields !!"
    
    # Configuración del modelo
    config = XttsConfig()
    config.load_json(xtts_config)
    
    # Inicializar el modelo
    XTTS_MODEL = Xtts.init_from_config(config)
    print("Loading XTTS model!")
    
    # Cargar el checkpoint del modelo
    XTTS_MODEL.load_checkpoint(config, checkpoint_path=xtts_checkpoint, vocab_path=xtts_vocab, use_deepspeed=False, weights_only=True)
    
    print("Model Loaded!")

# Función para ejecutar TTS
def run_tts(lang, tts_text, speaker_audio_file):
    if XTTS_MODEL is None or not speaker_audio_file:
        return "You need to run the previous step to load the model !!", None, None

    # Usar inference_mode para mejorar el rendimiento
    with torch.inference_mode():
        gpt_cond_latent, speaker_embedding = XTTS_MODEL.get_conditioning_latents(
            audio_path=speaker_audio_file, 
            gpt_cond_len=XTTS_MODEL.config.gpt_cond_len, 
            max_ref_length=XTTS_MODEL.config.max_ref_len, 
            sound_norm_refs=XTTS_MODEL.config.sound_norm_refs
        )
        
        out = XTTS_MODEL.inference(
            text=tts_text,
            language=lang,
            gpt_cond_latent=gpt_cond_latent,
            speaker_embedding=speaker_embedding,
            temperature=XTTS_MODEL.config.temperature,
            length_penalty=XTTS_MODEL.config.length_penalty,
            repetition_penalty=XTTS_MODEL.config.repetition_penalty,
            top_k=XTTS_MODEL.config.top_k,
            top_p=XTTS_MODEL.config.top_p,
        )
    
    # Guardar el audio generado en un archivo temporal
    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as fp:
        out["wav"] = torch.tensor(out["wav"]).unsqueeze(0)
        out_path = fp.name
        torchaudio.save(out_path, out["wav"], 24000)
    print("Speech generated!")

    return out_path, speaker_audio_file

# Definir la función para Gradio
@spaces.GPU
def generate(text, audio):
    load_model(model_path, config_path, vocab_path)
    out_path, speaker_audio_file = run_tts(lang='es', tts_text=text, speaker_audio_file=audio)
    return out_path

# Configurar la interfaz de Gradio
demo = gr.Interface(
    fn=generate, 
    inputs=[gr.Textbox(label='Frase a generar'), gr.Audio(type='filepath', label='Voz de referencia')], 
    outputs=gr.Audio(type='filepath')
)

# Lanzar la interfaz
demo.launch()