How to split a song into stems (vocals, drums, bass)

By Tongkit Leong · June 25, 2026 · 7 min read

Stem splitting separates a finished audio file into isolated parts — vocals, drums, bass, and everything else — using AI source separation. Tracks runs this extraction directly in your browser: import any audio file, run Extract Stems, and you get four separate tracks with no file upload, no account, and nothing stored anywhere but your own device.

What are stems?

In music production, a "stem" is an isolated layer of a finished mix. A full song has all the instruments and vocals combined into a single stereo file. Splitting it into stems means separating those elements so you can hear — and work with — each part on its own.

The four categories that stem extraction tools typically produce are:

• Vocals — all sung or spoken material
• Drums — kick, snare, hi-hats, and percussion
• Bass — bass guitar, 808s, bass synths
• Other — guitars, keys, strings, synth pads, and anything that does not fit cleanly into the other three categories

These are not necessarily clean separations — a bass guitar's harmonics overlap with a guitar's low end, and a melodic synth line might bleed between vocals and other — but modern source separation is good enough to be useful for most production and sampling tasks.

What people use stem splitting for

Stem splitting has practical uses at every skill level, from beginners learning songs to experienced producers building new tracks.

Remixing and sampling

The most common use is extracting elements from an existing track to build something new. You might pull the vocal from a song and set it over a different beat, or lift a drum loop from a record and repitch it. Having the isolated stem gives you a clean element to work with rather than a full mix where the drums are tangled with the bass and guitars.

Be aware that extracting stems from commercially released music does not grant you any rights to use those elements — stem splitting is a tool, not a licence. For original work or properly cleared samples it is entirely appropriate.

Removing vocals for practice or karaoke

If you want to sing or play along to a track, extracting and muting the vocal stem gives you an instrumental version. This is not perfect — some vocal frequencies will remain in the other stems — but it works well enough for practising against a familiar recording or creating a backing track.

Studying a mix

Isolating each stem lets you hear exactly what the original producer did with each element. How much reverb is on the snare alone? What does the bass actually sound like without the kick masking it? How is the lead vocal processed? Stem separation makes these details audible that are otherwise hidden in the full mix.

Fixing a bad export

If you finished a song but lost the project file, stem splitting your own export is sometimes the fastest route back to an editable state. The stems will not be as clean as your original tracks, but they give you something to work with rather than starting from scratch.

How stem extraction works

Source separation uses machine learning models trained on large sets of music with known stem labels. The model learns the statistical patterns of what drums sound like versus what vocals sound like, and at inference time it applies those learned patterns to separate a new audio file it has never seen before.

This happens entirely on your device in Tracks. The audio never leaves your browser — there is no server-side processing, no upload, and no account needed. Processing runs in the Web Audio API using your CPU, which is why it takes a few minutes for a full-length track rather than happening instantly.

The quality depends on the complexity of the original recording. Tracks with very clear separation between instruments (sparse electronic music, a solo vocal over piano) tend to separate cleanly. Dense mixes with many overlapping elements in the same frequency range (full-band recordings with lots of reverb) produce results with more bleed between stems.

How to split a song into stems in Tracks

The whole process takes five steps. Open the Tracks browser studio before you begin — no download or sign-in required.

Step 1: Import your audio file

Drag your audio file — MP3 or WAV — directly onto the timeline, or use the Import button at the top of the studio. The file will appear as a waveform clip on a new audio track. Press play to confirm it sounds correct before starting extraction.

If your file is very long (more than five minutes), consider trimming it to the section you actually need before extracting — extraction time scales with the length of the audio, and shorter files process faster.

Step 2: Select the clip and open Extract Stems

Click the clip to select it. In the clip's right-click menu or the track toolbar, choose Extract Stems. A panel will open showing the four output categories: Vocals, Drums, Bass, and Other.

Step 3: Run the extraction

Click Extract to start the separation. Tracks processes the audio locally — you will see a progress bar while the model works through the file. For a typical three-minute track, expect one to three minutes of processing time, depending on your device's CPU speed. Keep the browser tab active during processing; putting the tab in the background may slow it down on some devices.

Step 4: Review the stems in the timeline

When extraction finishes, four new audio tracks appear in the timeline — one for each stem. Use the Solo button on each track to listen to it in isolation. Check the vocal stem for bleed from instruments, and the drum stem for bass frequencies that may have leaked through. The Other stem often contains the widest range of elements, so give it a careful listen to hear what ended up there.

You can process each stem like any other audio track in Tracks — add EQ to clean up frequencies, apply compression, or use reverb sends. This is a good place to reduce low-end bleed on the vocal stem with a high-pass filter, or to tighten transients on the drum stem with gentle compression.

Step 5: Export the stems you want

To save a stem as a separate file, select its track and use Export Track to download it as a WAV or MP3. If you want to export a custom combination — for example, the drums and bass together without vocals — mute the tracks you do not want and use the full-project export. See How to export your song to MP3 or WAV for the full export workflow.

Getting the most from your stems

Clean up bleed with EQ

Even good source separation leaves some bleed. A high-pass filter at 100–150 Hz on the vocal stem removes low-end rumble that leaked through from the kick and bass. A low-pass filter at 8–10 kHz on the bass stem removes upper-frequency bleed from guitars or cymbals. These simple EQ cuts can make the stems feel much more separated than the raw output.

Use the stems as arrangement reference

If you are making a lo-fi remix or beat, bring the drum stem into Tracks and chop it into loops — select the segment you want, export it as a short clip, and import it as a sample. See Make lo-fi hip-hop in your browser for a walkthrough of building a track from chopped samples and recorded elements.

Time-stretch without re-extracting

If the stem is at a different tempo from your project, use Tracks' time-stretch and repitch controls on the audio clip to fit it to your grid. Select the clip, enable time-stretch mode, and drag the clip's edge to the new length. Tracks will stretch the audio to match without changing the pitch.

Frequently asked questions

Does my audio file get uploaded to a server?

No. Stem extraction in Tracks runs entirely on your device using the Web Audio API. The audio file never leaves your browser — there is no upload, no cloud processing, and nothing stored on an external server. Your projects and imported audio stay local only.

What audio file formats can I import for stem extraction?

Tracks accepts MP3 and WAV files for import. Both work with stem extraction. WAV files give the extraction model the cleanest signal to work with, since MP3 compression introduces artefacts that can affect separation quality slightly — though in practice the difference is usually small.

How good is the output quality?

Quality varies with the source material. Tracks with clear tonal separation — a vocal over sparse accompaniment, a simple drum machine pattern, a clean bass without distortion — separate well. Dense, reverb-heavy mixes with many layers in the same frequency range produce more bleed between stems. Treat the output as a useful starting point rather than a perfect separation, and apply EQ to refine it.

Can I extract stems from a track I recorded in Tracks?

Yes, but for your own Tracks projects, you do not need to. Export each track individually using Export Track rather than running stem extraction on a full mix — per-track export gives you a perfectly clean stem with no separation artefacts. Stem extraction is designed for audio files where you do not have the original multi-track project.

How long does extraction take?

Processing time depends on the length of the audio and your device's CPU. A three-minute track typically takes one to three minutes on a modern laptop. A seven-minute track may take four to six minutes. Keeping the browser tab active and foreground during processing speeds things up on most devices.

What is the "Other" stem?

The Other stem captures everything that the model does not classify as vocals, drums, or bass — typically guitars, keyboards, strings, synth pads, and other melodic or harmonic instruments. If you are extracting a guitar riff from a full-band recording, it will most likely appear in the Other stem.

Can I use the extracted stems in a new Tracks project?

Yes. After extraction, each stem is a regular audio track in the timeline. You can drag the clips, trim them, apply effects, or export them and import them into a new project. The studio treats extracted stems exactly like any other imported audio.