How to split stems from any song

Stem splitting, stem separation, source separation, and demixing all describe the same basic job: taking a finished stereo song and estimating the parts inside it. A four-stem splitter returns vocals, drums, bass, and other instruments as separate audio files. You can then mute the vocals for karaoke, isolate drums for sampling, study a bass line, or build a rough remix without the original studio session.

Modern separators use neural audio models because instruments overlap in both time and frequency. A voice and a guitar can occupy the same notes, so a simple EQ cannot cleanly pull one away from the other. The model looks for patterns it learned from mixed songs and their component tracks, then predicts the most likely audio for each stem. The result can be extremely useful, but it is still an estimate: some bleed, softened transients, or watery artifacts may remain.

Step 1: Open voicechanger.live/tools/stem-splitter on a desktop browser. There is no account or software installation. Choose an MP3, WAV, FLAC, OGG, or another audio file your browser can decode, up to 200 MB and 10 minutes.

Step 2: Start the four-stem separation. On the first run, the browser downloads and verifies the local separation model; later visits can reuse verified cached model data. Your song stays on your device. The model download is the network request, not an upload of your audio.

Step 3: Let the browser build vocals, drums, bass, and other instruments. Processing time depends on song length, arrangement, browser support, memory, and your computer. A short track on a fast GPU may finish quickly, while a long or dense song can take several minutes.

Step 4: Judge the result at the same moment. Compare the original with the stem mix, seek to difficult sections, then mute, solo, or change the level of each stem. Choruses, layered vocals, cymbals, reverb tails, and distorted guitars are useful stress tests.

Step 5: Download what you need. Save an individual stem as WAV, download all four WAV files in a ZIP, or save the audible level mix as a WAV. If you only need vocals plus instrumental, use the separate Vocal Remover instead of running a four-stem job.

Multitracks are the individual recordings captured during production: lead vocal, backing vocals, kick, snare, overheads, bass DI, guitars, synths, and more. Stems are grouped parts such as all drums or all vocals. When a browser stem splitter works from a final stereo master, its outputs are AI estimates rather than the original multitracks.

Use official multitracks or stems when you have them; they contain the cleanest separation and the producer's intended edits. AI-separated stems are valuable when the session files are unavailable and you need a practice track, remix sketch, sample idea, transcription aid, or DJ edit. Treat them as workable source material, not a replacement for a missing studio archive.

Choose the Stem Splitter when you need individual control over vocals, drums, bass, and the remaining instruments. That is the useful route for remixing, drum sampling, bass transcription, backing tracks, and arrangements where you want to mute or rebalance several parts.

Choose the Vocal Remover when the job is simply vocals versus instrumental. Its dedicated two-source model returns those two files directly, which is a better fit for karaoke, acapellas, and quick instrumental versions. The two tools own different jobs; the Stem Splitter does not hide a two-stem selector on the same page.

Start with source quality. A lossless WAV or FLAC can give the model more detail than a heavily compressed MP3, although arrangement difficulty usually matters more than the file extension alone. Sparse productions with clearly separated parts tend to be easier than dense mixes with doubled vocals, distorted guitars, wide reverbs, or instruments sharing the same range.

Listen for bleed: a hi-hat faintly present in the vocal, a vocal reverb tail inside the instruments, or bass harmonics leaking into the other stem. Then listen for artifacts such as fluttering cymbals, softened attacks, phasey ambience, or brief holes where two sources overlap. Compare the original and mix at the same playhead, solo each stem, and check both quiet verses and busy choruses before downloading.

For a remix or sample, trim around the cleanest phrase and mix the separated part under new material rather than expecting complete isolation. EQ, fades, and careful arrangement can hide small artifacts. A noise gate may help with steady low-level bleed, but aggressive cleanup can remove quiet notes and natural tails.

Remixing and mashups: Extract a vocal for a new instrumental, rebalance drums and bass, or mute the original arrangement while sketching a different one. The custom mix is useful for a rough reference before moving the WAV stems into a DAW.

Sampling and beat-making: Isolate a drum break, bass phrase, vocal hook, or melodic texture. Separation can reduce competing parts, but audition the sample in context and expect to trim around bleed rather than assuming every hit will be perfectly clean.

DJ preparation: Build an acapella, instrumental, or reduced backing mix for transitions and live mashups. Check the loudest chorus and any reverb-heavy outro before relying on a stem in a set.

Practice and transcription: Solo the bass, drums, vocal, or remaining instruments to learn a part by ear. Mute your instrument and play along with the rest of the song, or export a backing mix for rehearsal.

Karaoke: If you only need vocals removed, the Vocal Remover is the direct two-source route. Use four-stem separation when you also want to reduce drums, bass, or other instruments for a custom backing track.

The tool changes the audio format; it does not give you rights to the song or its separated parts. Use music you created, public-domain material, appropriately licensed tracks, or audio you otherwise have permission to process and reuse.

Rules for samples, remixes, performances, education, private study, and commercial releases vary by work, license, platform, and jurisdiction. A separated stem can still contain copyrighted music. If you plan to publish, distribute, monetize, or perform the result, check the relevant permissions and seek qualified advice when the rights are unclear.