What Is Stem Separation?
Stem separation — also called source separation or audio demixing — is the process of splitting a mixed song into its individual component tracks. A finished song is a blend of dozens of sounds recorded and mixed together. Stem separation works backward: it uses AI to untangle that blend and hand you back each layer independently.
The standard output is four stems:
Modern AI separation uses the Demucs model from Meta Research, trained on thousands of hours of professionally mixed music. It outperforms older methods like spectral masking by a significant margin — especially on complex, dense productions.
Drum Sub-Separation StemForge Exclusive
Standard stem separation gives you one drum stem — the whole kit mixed together. StemForge goes further. After separating the main stems, you can trigger drum sub-separation: a second AI pass that breaks the drum stem into 6 individual components.
This level of separation is not available on Moises, LALAL.AI, or any other free online tool. It uses the SFXMR model, specifically trained on drum source separation. Producers use drum sub-separation to sample individual hits, build MIDI drum maps, analyze groove patterns, or simply study how a professional drummer played a kit.
To use it: complete a standard 4-stem separation, then click the 🥁 Separate button that appears on the drums card. Results appear below the drums stem in 2–4 minutes.
How to Separate Stems with StemForge
Three steps. No account, no watermarks, no credit card.
Upload your track
Drag and drop your file into the upload zone at StemForge, or click to browse. Accepted formats: WAV, MP3, FLAC, M4A. Maximum file size: 50MB. A waveform preview appears once your file is loaded.
Click "Separate Stems"
Hit the orange button to start processing. A live progress bar shows separation status. Most tracks complete in 1–3 minutes depending on length and server load. You can play a demo track while you wait to see what the output looks like.
Play, preview, and download
Four stem rows appear — vocals, drums, bass, other — each with an individual waveform, playback controls, and a download button. Download any stem as a WAV file. Optionally click 🥁 on the drums card to trigger 6-stem drum sub-separation.
Common Use Cases
What producers, musicians, and creators actually do with separated stems:
🎵 Remixing
Pull vocal acapellas and instrumental beds from released tracks. Rearrange and rework songs in your DAW without needing the original session files.
🎤 Karaoke tracks
Remove vocals to create backing tracks for live performance, rehearsal, or karaoke events. More accurate than older vocal remover tools.
🎸 Practice along
Guitarists mute the guitar stem and play along to bass and drums. Bassists do the same. Useful for learning songs and testing your ear.
🌷 Sampling
Extract drum hits, bass grooves, or vocal chops for use in new productions. Drum sub-separation gives you isolated kicks and snares ready to sample.
🎵 Transcription
Isolate a bass line or melody to transcribe it by ear. Removing competing frequencies makes it significantly easier to hear individual parts.
🎙 Stem mastering
Separate a stereo mix into 4 stems and master each layer independently before recombining — an alternative to sending full session files to a mastering engineer.
Tips for Best Results
Use high-quality source files. Lossless formats (WAV, FLAC) produce noticeably better stem separation than heavily compressed MP3s. If you have 320kbps MP3, that works. Avoid low-bitrate files (<128kbps) — the compression artifacts bleed into the stems.
Avoid reverberated or live recordings. Room reverb smears the frequency spectrum and makes it harder for the model to isolate sources. Studio recordings with dry, close-miked instruments separate most cleanly.
Dense electronic music is harder. Demucs was primarily trained on acoustic instruments. Heavily layered synth pads, sidechain-compressed bass, and synthesized drums may bleed between stems. Organic instruments (guitar, piano, acoustic drum kit) separate with higher accuracy.
For drum sub-separation, run main separation first. The 6-stem drum breakdown processes the isolated drum stem, not the full mix. Starting from a clean drum stem (without vocal and instrument bleed) gives you cleaner individual drum components.
Frequently Asked Questions
Do I need an account to use StemForge?
No. Upload, separate, and download stems without creating an account or providing an email address. The free tier allows 3 separations per day per IP address.
What audio formats are supported?
WAV, MP3, FLAC, and M4A up to 50MB. For best separation quality, use uncompressed WAV or FLAC at 44.1kHz or 48kHz.
How long does separation take?
Standard 4-stem separation: 1–3 minutes depending on track length. Drum sub-separation: an additional 2–4 minutes after the main pass.
Can I use separated stems commercially?
StemForge does not place restrictions on your use of the output files. Copyright ownership of separated stems remains with the original copyright holder — ensure you have rights to the source material before commercial use.
What AI model does StemForge use?
4-stem separation uses Demucs (Meta Research), the leading open-source model for music source separation. Drum sub-separation uses the SFXMR model trained specifically for drum component isolation.
How does StemForge compare to Moises or LALAL.AI?
StemForge is free (3/day, no login), runs the same Demucs model as paid tools, and adds 6-stem drum sub-separation that no competitor offers. See the full comparison →