app.py

import os
import gradio as gr
from scipy.io.wavfile import write
import spaces
import demucs.separate

@spaces.GPU()
def inference(audio):
  os.makedirs("out", exist_ok=True)
  write('test.wav', audio[0], audio[1])
  demucs.separate.main(["-n", "htdemucs", "--two-stems=vocals", "test.wav", "-o", "out"])
  return "./out/htdemucs/test/vocals.wav", "./out/htdemucs/test/no_vocals.wav"

title = "Demucs Music Source Separation (v4)"
article = "<p style='text-align: center'><a href='https://arxiv.org/abs/1911.13254' target='_blank'>Music Source Separation in the Waveform Domain</a> | <a href='https://github.com/facebookresearch/demucs' target='_blank'>Github Repo</a> | <a href='https://www.thafx.com' target='_blank'>//THAFX</a></p>"

with gr.Blocks(title=title) as demo:
    gr.Markdown(f"# {title}")
    with gr.Row():
        audio_input = gr.Audio(type="numpy", label="Input")
    with gr.Row():
        vocals_output = gr.Audio(type="filepath", label="Vocals")
        no_vocals_output = gr.Audio(type="filepath", label="No Vocals / Instrumental")
    with gr.Row():
        submit_btn = gr.Button("Submit", variant="primary")
        clear_btn = gr.ClearButton([audio_input, vocals_output, no_vocals_output])
    submit_btn.click(fn=inference, inputs=audio_input, outputs=[vocals_output, no_vocals_output], api_name="predict")
    gr.HTML(article)

demo.queue().launch()