demix 1.3.0

Separate audio into stems (vocals, instruments) using AI-powered Spleeter

demix

separates audio from songs into stems (vocals, instruments), transposes music, changes tempo, detects music key, operates on music from youtube video or local file

prerequisites

[!NOTE] Please note: I'm using homebrew for installing virtualenvwrapper and ffmpeg. If you're using another package manager or different operating system than macOS (e.g. Linux), you need to install it differently. In the examples for creating virtualenv below, replace python path below with your own python3.8 path. I'm personally using pyenv for installing various python verions, but you can do it as you wish.

installation

I suggest to create virtualenv for this project to not break existing system-wide installations:

brew install virtualenvwrapper
brew install ffmpeg
mkdir demix
cd demix
mkvirtualenv -p /Users/pw/.pyenv/versions/3.8.16/bin/python demix
workon demix
pip install demix
demix -v

update

pip install demix --upgrade

development

prepare environment:

brew install virtualenvwrapper
brew install ffmpeg
git clone [e-mail]:pwittchen/demix.git
cd demix
mkvirtualenv -p /Users/pw/.pyenv/versions/3.8.16/bin/python demix
workon demix
pip install -r requirements.txt
python demix.py -v

virtualenv

exit virtualenv, when you're done:

deactivate

to activate env again:

workon demix

testing

install pytest:

pip install pytest

run all tests (-v param for verbose):

pytest -v

versioning and deployment

When we create and push a new git tag, e.g. v1.0.4, deploy.yml github action is triggered. It automatically extracts created tag, updates version with bump_version.py script, performs git commit and push. After that, deployment of the new package version to PyPi is executed.

usage

demix -u <youtube-url> [options]
demix -s <search-query> [options]
demix -f <audio-file> [options]

options

Option	Description
-u, --url	YouTube video URL to process
-s, --search	Search YouTube for a song (e.g., 'Artist - Song Name')
-f, --file	Local audio file to process (mp3, wav, flac, etc.)
-o, --output	Output directory (default: output)
-t, --tempo	Tempo factor for output audio (default: 1.0, use < 1.0 to slow down)
-p, --transpose	Transpose pitch by semitones (default: 0, range: -12 to +12)
-k, --key	Detect and display the musical key of the audio
-K, --target-key	Transpose audio to target key (e.g., C, Am, F#, Bb minor)
-ss, --start	Start time for cutting (format: MM:SS or HH:MM:SS)
-to, --end	End time for cutting (format: MM:SS or HH:MM:SS)
-m, --mode	Separation mode: 2stems, 4stems, or 5stems (default: 2stems)
-c, --clean	Clean up files: output, models, or all
-v, --version	Show version number
-h, --help	Show help message

separation modes

Mode	Stems
2stems	vocals, accompaniment
4stems	vocals, drums, bass, other
5stems	vocals, drums, bass, piano, other

examples

# separate a YouTube video into vocals and accompaniment
demix -u 'https://www.youtube.com/watch?v=VIDEO_ID'

# search YouTube by artist and song name
demix -s 'Queen - Bohemian Rhapsody' -m 4stems

# separate a local file with 4 stems
demix -f /path/to/song.mp3 -m 4stems

# cut audio from 1:30 to 3:45 before separation
demix -f song.mp3 -ss 1:30 -to 3:45

# start from 0:30 (skip intro)
demix -f song.mp3 -ss 0:30

# keep only the first 2 minutes
demix -f song.mp3 -to 2:00

# combine cutting with tempo and transpose
demix -f song.mp3 -ss 1:00 -to 4:00 -t 0.8 -p -2

# detect musical key of a song
demix -f song.mp3 -k

# detect key before and after transposing
demix -f song.mp3 -k -p -3

# transpose to a specific key (auto-detects current key)
demix -f song.mp3 -K C

# transpose to A minor
demix -f song.mp3 -K Am

# transpose to target key with tempo change
demix -f song.mp3 -K "F# minor" -t 0.9