app.py

import torch
import torchaudio
import gradio as gr
import os
from demucs.pretrained import get_model
from demucs.apply import apply_model

# Load mdx_extra model
model = get_model('mdx_extra')
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

output_dir = "separated_mdx_extra"
os.makedirs(output_dir, exist_ok=True)

def separate_audio(audio_path):
    wav, sr = torchaudio.load(audio_path)
    wav = wav.to(device)

    sources = apply_model(model, wav[None], device=device)

    sources_dict = {}
    stems = ["vocals", "instrumental"]

    for source, stem in zip(sources[0], stems):
        stem_path = os.path.join(output_dir, f"{stem}.wav")
        torchaudio.save(stem_path, source.cpu(), sr)
        sources_dict[stem] = stem_path

    return sources_dict["vocals"], sources_dict["instrumental"]

# Gradio Interface
interface = gr.Interface(
    fn=separate_audio,
    inputs=gr.Audio(type="filepath"),
    outputs=[gr.Audio(label="Vocals"), gr.Audio(label="Instrumental")],
    title="AI Music Separator (Demucs)",
    description="Upload a song, and AI will separate vocals and instrumental."
)

if __name__ == "__main__":
    interface.launch()