Every audio file in the folder becomes a track with its own volume, pan, mute and solo. Mute your own part and play along.
Files are named like Artist - Song - bass.wav; the last part becomes the track name.
Your browser may ask to confirm an “upload”. Nothing leaves this computer; the files are only read locally.
0:00.0 / 0:00.0
ZOOM
MASTER
TEMPO
—
NUDGE0 ms
BEATS / BAR
COUNT-IN
PLAY RANGEWhole song
◂ drag along this bar to set a play range ▸
Space play / pause · ←→ skip 5 s · Home start · R repeat · IO range in / out · click the timeline to jump · click a track name to rename it · double-click a fader or knob to reset
GUIDE
STEM◆MIXER · QUICK GUIDE
How to use Stem Mixer
Ten things to know. The numbers on the picture match the list.
1234566788910
1
Open a song Press Open stem folder and pick the song's folder. If the browser says “upload”, nothing leaves your computer.
2
Play Press Play or the space bar. The left and right arrow keys skip 5 seconds.
3
Count-in Pick how many clicks you get before the music starts: 1 to 8, or Off.
4
Mute your part MUTE silences a track so you can play that part yourself. SOLO plays only that track.
5
Balance the mix Every track has a volume slider and a left/right pan knob. Double-click either one to reset it.
6
Loop a hard section Drag along the bar above the tracks to mark a section, then switch on Repeat to hear it again and again.
7
Zoom Zoom in to see the sound up close and place a loop exactly. FIT shows the whole song.
8
Click track Unmute CLICK for a metronome that follows the recording. Ticking twice too fast or too slow? Press ÷2 or ×2.
9
Rename a track Click a track's name to change it, for example OTHER to KEYS. The files themselves are not changed.
10
Save your setup Save settings keeps your mix for this song. Next time you open it, the page offers to bring it back.
STEM◆MIXER · HOW IT WORKS
From playlist to practice
How a Spotify playlist turns into songs you can play along with, one instrument at a time.
ONE PERSON DOES THIS, ONCE PER SETLIST
STEP 1
Pick the songs
Build a Spotify playlist of the covers the band is learning.
STEP 2
Find each song
Give the playlist to Claude. It finds the studio version of every track on YouTube and lists the links.
STEP 3
Save the audio
A free download tool called yt-dlp saves the sound of each video as an MP3 file.
STEP 4
Split the band apart
StemRoller, a free app, separates each song into its parts using its 6-channel setting.
EVERYONE IN THE BAND
STEP 5
Practice
Open the song's folder in Stem Mixer, mute your own part and play along.
WHAT STEP 4 MAKES FOR EVERY SONG
VOCALS
DRUMS
BASS
GUITAR
PIANO
OTHER
INSTRUMENTAL
ORIGINAL
Each part is its own audio file. In Stem Mixer every file becomes a track with its own volume, pan, mute and solo. Instrumental and Original are the whole song, so they start muted.
Good to know. The split is done by software, so a little of one instrument can leak into another. If a track has the wrong label, click its name in the mixer and rename it.
STEM◆MIXER · PIPELINE, FOR ENGINEERS
Playlist → stems → browser mixer
Five stages, mostly off-the-shelf tools. The only custom code is the last stage: one static HTML file.
STAGE 1 · INPUT
Spotify playlist
A playlist URL. It defines the ordered track list: artist and title for every song.
OUT → track list
STAGE 2 · RESOLVE
Claude + web search
Looks up the studio version of each track on YouTube. Returns one watch URL per track, which gets a quick human check.
OUT → watch URLs
STAGE 3 · FETCH
yt-dlp in Docker
Extracts audio only and converts it to MP3 with ffmpeg at best VBR quality. One file per track, numbered in playlist order.
OUT → MP3 files
STAGE 4 · SEPARATE
StemRoller
Demucs-based desktop app. The 6-channel model writes vocals, drums, bass, guitar, piano and other, plus instrumental and original, as stereo WAV.
OUT → WAV stems
STAGE 5 · MIX
stem-mixer.html
One static HTML file. No server, no build, no dependencies. Reads the stem folder locally and plays it with the Web Audio API.
OUT → speakers
STAGE 2 OUTPUT · URL LIST# one line per track, in playlist order
https://www.youtube.com/watch?v=<id of track 1>
https://www.youtube.com/watch?v=<id of track 2>
https://www.youtube.com/watch?v=<id of track 3>
…
STAGE 3 · ONE COMMAND PER TRACKyt-dl -x --audio-format mp3 --audio-quality 0 \
-o "NN - Artist - Title.%(ext)s" "<watch URL>"
# yt-dl is a shell alias for docker run, with
# $(pwd) mounted at /workdir. The image holds
# yt-dlp, ffmpeg and Deno, the JS runtime that
# yt-dlp needs for YouTube.
STAGE 4 OUTPUT · ONE FOLDER PER SONGArtist - Title/
Artist - Title - vocals.wav
Artist - Title - drums.wav
Artist - Title - bass.wav
… guitar, piano, other,
instrumental, original
INSIDE STAGE 5 · WHAT THE PAGE DOES WITH A STEM FOLDER
Folder picker
A webkitdirectory file input returns File objects. Nothing is uploaded.
Decode
decodeAudioData turns every file into an AudioBuffer in memory.
Sources
One buffer source node per stem, all started at the same clock time.
Gain
Per-track fader. Mute and solo drive this gain to zero.
StereoPanner
Per-track pan, from hard left to hard right.
Master gain
Sums every track and feeds the speakers.
ALONGSIDE PLAYBACK
Waveforms
Min and max peaks are precomputed at several resolutions and drawn on a canvas, only for the visible part of the timeline.
Beat tracker
Onset envelope from the drum stem, a tempo path that is allowed to drift, then dynamic-programming beat tracking. Runs once per song.
Click and count-in
Clicks are short buffers scheduled on the audio clock along the beat grid. Loops are scheduled a few seconds ahead, pass by pass.
Saved settings
Volume, pan, mute, solo, track names, play range and tempo fixes go in a cookie or localStorage, keyed by a hash of the file names.