Upload a song and the AI will separate it into 4 individual stems: vocals, drums, bass, and other instruments. Use the solo and mute controls to preview any combination, adjust per-stem volume, then download each stem individually or grab all 4 at once as a ZIP file. The same state-of-the-art AI that powers our Vocal Remover — completely free, unlimited, and your files never leave your device.
Max size: 100 MB · Max duration: 10 minutes
A typical 3–4 minute song takes 10–15 minutes in Chrome and 20–30 minutes in Safari. Newer machines (especially Apple M-series) are significantly faster.
The progress bar shows a live estimate — close other browser tabs for best performance. Grab a coffee while it works!
What Is AI Stem Splitting?
Stem splitting (also called music source separation or music demixing) is the process of taking a fully mixed song and separating it into its individual component tracks — called stems. This tool uses Demucs v4, Meta Research's state-of-the-art deep learning model that won the Sony Music Demixing Challenge, to separate your song into four high-quality stems: vocals, drums, bass, and other instruments.
Unlike older methods that relied on simple frequency filtering (which produced terrible artifacts and only worked in specific conditions), Demucs v4 is a neural network that has learned what each instrument actually sounds like — so it can identify and isolate them even when they overlap in frequency, time, and space within the stereo field. The result is dramatically better separation than any non-AI approach.
Because the AI runs directly in your browser using WebAssembly, your music files never leave your device. No data is uploaded to any server. This is a real privacy advantage over cloud-based competitors like lalal.ai, moises.ai, and LALAL.AI that require uploading your audio.
Use Cases for Stem Splitting
Music Production & Remixing: Extract clean stems from any song to use in your own productions. Isolate a drum groove to sample, grab a bass line to build on, or pull the vocal for a remix or mashup. Each stem downloads as a full-resolution WAV file, ready to drop straight into Ableton, FL Studio, Logic, Pro Tools, or any DAW.
Learning & Practice: Isolate any instrument to study it in detail. Slow down the bass stem to learn the exact notes, loop the drum stem to practice over it, or mute the melody to test your ear training. Great for music students, musicians learning new songs by ear, and anyone studying arrangement and production techniques.
Backing Tracks & Karaoke: Create custom practice backing tracks by muting specific instruments. Mute the bass to practice bass guitar. Mute the drums to practice drumming. Mute the guitar to practice soloing. This is more flexible than a simple vocal remover because you can exclude any instrument combination — not just the vocals.
DJ & Live Performance: Create stems for DJ sets, mashups, and live remixes. Swap the drums from one song into another, layer vocal acapellas over new instrumentals, or build multi-stem live sets. The synchronized playback lets you audition any combination before you download.
Sound Design & Sampling: Extract individual elements — a snare hit, a bass note, a vocal chop — for use as samples in new productions. Clean stems are far more usable for sampling than trying to chop directly from a fully mixed song.
How to Split a Song Into Stems — 3 Steps
Step 1: Upload Your Song
Click the upload area or drag and drop any audio file (MP3, WAV, FLAC, OGG, M4A, AIFF, or WEBM). The maximum file size is 100 MB and the maximum duration is 10 minutes. On your first use, there's a one-time setup download of approximately 80 MB (the AI model weights). This takes 15–60 seconds depending on your connection and is cached in your browser so future uses load instantly.
Step 2: AI Stem Separation
The AI analyzes your song and separates it into 4 individual stems. A progress bar shows the percentage complete and estimated time remaining. Processing typically takes 10–20 minutes for a standard-length song on a modern desktop computer. The AI is doing substantial work — it's running a 75-million parameter neural network entirely in your browser, with no server involved.
Step 3: Mix, Preview & Download
When processing is complete, all four stems appear with a synchronized master transport. Press play to hear all stems together, then use the Solo and Mute buttons on each stem to preview any combination. Adjust per-stem volume with the volume sliders. When you're satisfied, download individual stems as WAV or MP3, or click Download All as ZIP to get all 4 stems in one package.
Understanding the 4 Stems
🎤 Vocals: The isolated singing and vocal performance, completely separated from all instruments. This is the same output as our Vocal Remover's isolated vocals track. Use it for acapellas, vocal remixes, studying a singer's phrasing and technique, or creating custom karaoke-style tracks where you control which other instruments are present.
🥁 Drums: The complete drum kit — kick drum, snare, hi-hats, ride and crash cymbals, toms, and any other percussion. Demucs v4 achieves approximately 10 dB SDR on drums, making it one of the most accurately separated stems. Great for extracting drum grooves, learning drum parts, or sampling drum patterns.
🎸 Bass: The bass guitar, electric bass, upright bass, sub-bass synthesizer, and any other dominant low-frequency melodic instruments. This stem is particularly useful for musicians learning bass lines, producers building new arrangements on top of an existing bass groove, or DJs who want to control low-end separately.
🎹 Other Instruments: Everything that doesn't fall into the above three categories — guitars (electric, acoustic, classical), synthesizers, piano, organ, strings, brass, woodwinds, and any other melodic or harmonic instruments. This is sometimes labeled "accompaniment" in research literature. The quality varies more across this stem since it's a catch-all category, but for most well-produced pop and rock recordings it produces very usable results.
Tips for the Best Stem Separation Results
Use high-quality source audio. The AI performs best on professionally recorded and mixed studio tracks. High-bitrate MP3s (192kbps+) or lossless audio (WAV, FLAC) will produce better results than heavily compressed 128kbps files. The quality of the separation is ultimately limited by the quality of the input.
Results vary by genre. Pop, rock, R&B, hip-hop, and most commercially produced music will produce excellent results because the stems are clearly separated in the original mix. Heavily layered experimental music, lo-fi recordings, live recordings with bleed between microphones, or genres where instruments occupy the same frequency range (like jazz with piano and guitar) may produce more artifacts.
Use Chrome or Edge on a desktop for best speed. These browsers allocate 8 GB of memory per tab and fully support multi-threaded WebAssembly, which means your device's full CPU power is used for processing. Firefox also works well. Safari works but allocates less memory (4 GB) and is noticeably slower.
Trim long tracks first. Songs over 7 minutes take significantly more memory and time to process. If you need stems from a specific section of a longer track, use our Audio Trimmer to cut it down first — this will speed up processing and reduce memory usage.
Close other browser tabs. The AI uses substantial CPU and RAM. Closing other tabs gives the processor more headroom and can significantly reduce processing time, especially on devices with limited memory.
Frequently Asked Questions
What is a stem splitter?
A stem splitter separates a mixed song into individual component tracks (stems). This tool separates any song into 4 stems: vocals, drums, bass, and other instruments.
How is this different from a vocal remover?
A vocal remover creates two outputs: instrumental (everything except vocals) and isolated vocals. A stem splitter goes further — it also separates the drums, bass, and other instruments from each other, giving you 4 individual stems total. Use the Vocal Remover if you just need karaoke or acapella. Use this Stem Splitter if you need individual instrument tracks for production, remixing, or learning.
Is this really free?
Yes, 100% free with no limits, no account, no watermarks, no subscriptions. Process as many songs as you want.
Do my files get uploaded anywhere?
Never. All processing happens in your browser. Your audio files never leave your device. This is a real privacy advantage over server-based tools.
How long does processing take?
A typical 3–4 minute song takes 10–20 minutes. Newer computers and those with Apple M-series chips are faster. Processing 4 stems takes roughly the same time as our vocal remover since both use the same underlying model.
Can I use the stems in my own music?
Technically yes — but you are responsible for copyright. For personal use, practice, and learning, using stems from commercially released music is generally considered fair use. For commercial releases or public performances, you'd need to obtain licenses from the rights holders.
What is the best output format — WAV or MP3?
Download as WAV if you plan to use the stems in a DAW for production — WAV is lossless and maintains full audio quality with no artifacts. Download as MP3 if you just need to listen to or share the stems and want a smaller file size.
Can I download all 4 stems at once?
Yes. Use the "Download All as ZIP" button at the bottom of the results section. This packages all 4 stems in your chosen format (WAV or MP3) into a single ZIP file.
My browser ran out of memory. What do I do?
Close other tabs and applications, then try again. If the problem persists, try a shorter clip using our Audio Trimmer to cut it to under 5 minutes. Using Chrome or Edge instead of Safari also helps — they allocate more memory per tab.
Does this work on iPhone or iPad?
No. iOS devices cannot run this tool due to memory limits enforced by Apple on all iOS browsers. Use a desktop or laptop computer instead.
Will you add a 6-stem model (guitar + piano)?
This is on our roadmap. Demucs has a 6-stem variant that also separates guitar and piano as individual stems. We plan to add this as an optional mode in a future update.
Stem Splitter vs. the Competition
SoundTools vs. lalal.ai: lalal.ai is server-based, requiring you to upload your audio files to their servers. Their free tier is heavily limited (short clips, few per day). Their paid plans run $15–$25/month. SoundTools is completely free, unlimited, and processes everything locally — your files never leave your device.
SoundTools vs. moises.ai: moises.ai offers stem splitting as part of a freemium subscription. The free tier has strict limits and many features are paywalled. SoundTools provides the same core stem separation (using an equally high-quality model) with no limits and no cost, ever.
SoundTools vs. freemusicdemixer.com: freemusicdemixer.com is the closest competitor in terms of client-side architecture. Their free tier is limited; the full tool requires a subscription. We use the same underlying Demucs v4 technology and offer equivalent quality — completely free.
The Privacy Advantage: Every server-based competitor stores your uploaded audio files on their servers (at least temporarily). If privacy matters — for unreleased music, demo recordings, or simply personal preference — client-side processing is the only truly private option. SoundTools processes everything in your browser with zero server involvement.