How to split stems from any song

Stem splitting (also called source separation) is the process of isolating individual instruments from a mixed audio track. Upload a finished song and the AI extracts separate files for vocals, drums, bass, and other instruments. Each "stem" can be used independently — mute the vocals for a karaoke track, isolate the drums for sampling, or extract the bass line for a remix.

This technology was impossible just a few years ago. Traditional audio processing could only boost or cut frequency ranges, which always left artifacts and bleed between instruments. Modern AI models (like Meta's Demucs and ByteDance's HTDemucs) use deep neural networks trained on millions of songs to intelligently separate overlapping sounds — even when instruments share the same frequency range.

Step 1: Open voicechanger.live/tools/stem-splitter in your browser. No download, no account, no installation required.

Step 2: Upload your audio file. The tool supports MP3, WAV, FLAC, and OGG formats. File size limit is 50MB (roughly 10 minutes of audio at standard quality).

Step 3: Select your separation mode. "4-stem" separates into vocals, drums, bass, and other instruments. "2-stem" separates into vocals and instrumental only (faster processing, useful when you just need a karaoke track).

Step 4: Wait for processing. The AI analyzes the entire track and generates separated stems. Processing time varies by track length — a 3-minute song typically takes 30-60 seconds. All processing happens in your browser using WebAssembly — your audio is never uploaded to any server.

Step 5: Download your stems. Each stem is available as a separate WAV file. Play them back individually to verify quality, then download the ones you need.

Remixing: Extract the vocal from a song and place it over your own instrumental. This is the foundation of remix culture — and AI stem splitting makes it accessible to anyone, not just producers with access to the original multitrack recordings.

Sampling: Isolate a drum break, bass line, or melodic phrase from any recording. Clean stems mean cleaner samples — no bleed from other instruments contaminating your chops.

DJ mashups: Create acapellas and instrumentals from any track for live DJ sets. Blend vocals from one song over the instrumental of another. Previously this required finding official acapella releases — now you can create them from any song.

Practice and learning: Isolate the guitar part from a song to learn it by ear. Mute the drums and play along with the rest of the band. Music teachers use stem splitting to create custom practice tracks for students.

Karaoke: Remove the vocals from any song to create an instant karaoke track. The AI separation is clean enough for casual karaoke — no vocal bleed or artifact issues that plagued older vocal removal tools.

Source quality matters: Higher quality input produces better separation. Use lossless formats (WAV, FLAC) when available. Heavily compressed MP3 files (128kbps or lower) produce noticeably worse results because the compression has already destroyed audio detail that the AI needs for separation.

Song complexity: Simple arrangements (vocals + guitar, electronic music with clean synthesis) separate better than dense orchestral arrangements or heavily layered production. The AI handles most modern pop, rock, hip-hop, and electronic music very well.

Post-processing: The separated stems may have subtle artifacts — slight reverb tails, quiet bleed from other instruments. For professional use, apply a noise gate to clean up the quiet sections and use EQ to remove any remaining bleed frequencies.

Vocal removal (voicechanger.live/tools/vocal-remover) is a subset of stem splitting — it extracts just the vocals and instrumental. If you only need a karaoke track or an acapella, vocal removal is faster because it runs a simpler 2-stem model.

Full stem splitting gives you 4 or more individual stems, which is essential for production work. Use vocal removal for quick karaoke tracks and stem splitting for serious music production, remixing, and sampling.