app.py

import os
import uuid
import torch
import re
import spaces
import gradio as gr
import torchaudio
from audiocraft.models import MusicGen
from audiocraft.data.audio import audio_write

# Configuración del dispositivo
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

# Configurar ZERO_GPU_PATCH_TORCH_DEVICE
ZERO_GPU_PATCH_TORCH_DEVICE = 1

# Cargar el modelo `musicgen-melody` una única vez
model = MusicGen.get_pretrained("facebook/musicgen-melody")

@spaces.GPU
def generate_music(description, melody_audio, duration):
    description = clean_text(description)
    model.set_generation_params(duration=int(duration * 1000))  # Convertir segundos a milisegundos

    try:
        with torch.no_grad():
            if description:
                description = [description]
                if melody_audio:
                    # Cargar el archivo de audio para remixar
                    melody, sr = torchaudio.load(melody_audio, normalize=True)
                    melody = melody.to(device) if torch.cuda.is_available() else melody
                    wav = model.generate_with_chroma(description, melody[None], sr)
                else:
                    wav = model.generate(description)
            else:
                wav = model.generate_unconditional(1)

            filename = f'{str(uuid.uuid4())}.wav'
            path = audio_write(filename, wav[0].cpu().to(torch.float32), model.sample_rate, strategy="loudness", loudness_compressor=True)

            if not os.path.exists(path):
                raise ValueError(f'Failed to save audio to {path}')
            
            return path

    except Exception as e:
        return str(e)

def clean_text(text):
    text = re.sub(r'http\S+', '', text)
    text = re.sub(r'[^a-zA-Z0-9\s]', '', text)
    return text

# Definir la interfaz de Gradio
description = gr.Textbox(label="Description", placeholder="Acoustic, guitar, melody, trap, D minor, 90 bpm")
melody_audio = gr.Audio(label="Melody Audio (optional)", type="filepath")
duration = gr.Number(label="Duration (seconds)", value=10, precision=0)
output_path = gr.File(label="Generated Music")

gr.Interface(
    fn=generate_music,
    inputs=[description, melody_audio, duration],
    outputs=output_path,
    title="MusicGen Melody Demo",
    description="Generate music using the MusicGen melody model. Optionally remix with an audio file. Download the generated audio file.",
    examples=[
        ["happy rock", None, 8],
        ["energetic EDM", None, 8],
        ["chillwave", "./assets/example_melody.mp3", 10]
    ]
).launch()