python 3 script to extract vocals from a song using the Docker researchdeezer/spleeter image

extract_vocals.py

#!/usr/bin/env python3

# Example:
# > cd /tmp
# extract_vocals.py '/Users/ypujante/Music/iTunes/iTunes Music/Music/Janet Jackson/Janet/08 This Time.m4a'
# Generates /tmp/08 This Time/vocals.wav and /tmp/08 This Time/accompaniment.wav

import argparse
import os
import subprocess

spleeter_docker_image = "researchdeezer/spleeter:3.7-2stems"

parser = argparse.ArgumentParser(usage='extract_vocals.py [-h] [-o output] <input>')
parser.add_argument("-n", "--dry-run", help="Dry run (prints what it is going to do)", action="store_true", dest="dry_run")
parser.add_argument("-v", "--verbose", help="Verbose", action="store_true")
parser.add_argument("-o", "--output", help="Output directory (default to current dir)")
parser.add_argument("input", help="Input file to extract vocals from")

args = parser.parse_args()

input = os.path.realpath(args.input)
output = os.path.realpath(args.output if args.output else os.getcwd())

dir, filename = os.path.split(input)

verbose = [] if not args.verbose else ['--verbose']

cmd = [
    'docker', 'run',
    '-v', f'{dir}:/input',
    '-v', f'{output}:/output',
    spleeter_docker_image,
    'separate', *verbose, '-i', f'/input/{filename}', '-o', '/output'
    ]

if args.dry_run:
    print(' '.join(cmd))
else:
    subprocess.run(cmd)

It works on macOS but you need to make sure that the Docker VM image has enough memory. If you see this kind of warnings:

2020-10-18 18:38:12.587523: W tensorflow/core/framework/cpu_allocator_impl.cc:81] Allocation of 467008080 exceeds 10% of system memory.

you should increase the size of the memory allocated to the VM