How to Isolate an Instrument in a Song

You can isolate an instrument from a song in a few minutes using the free AI stem splitter at Vocal Cut — no account, no upload, and no software to install. Drop in a track, let the model split it into separate parts, then solo the one you want and save it. How clean the result is depends on the instrument and how busy the mix is, and this guide is honest about where that line falls.

Which instruments can you actually isolate?

This is the part most tutorials skip. The AI model behind the tool (HTDemucs) always produces exactly four stems — a stem is one isolated layer of a mix — and those four are fixed:

• Vocals
• Drums
• Bass
• Other

Two of those map to a single instrument, which makes them the easy wins:

• Bass has its own dedicated stem, so pulling out a bassline is clean and reliable.
• Drums also get their own stem, so isolating the full kit is just as straightforward.

Everything else — guitar, piano and keys, synths, strings, brass and horns, organ — lands together in the "other" stem. There is no per-instrument "guitar" or "piano" output, because the model was never trained to produce one. So isolating a single melodic instrument really means soloing "other" and hearing what it contains. If you want the full picture of how these four layers work, see what audio stems are.

That "other" stem behaves very differently depending on the song:

• On a sparse mix — say vocals, one guitar, and light drums — "other" is essentially just that guitar. Clean result.
• On a dense mix with layered guitars, keys, and strings all at once, "other" bundles them together and you will hear more than one instrument. That is a real limitation, not a bug, and no free (or paid) tool fully separates instruments that were never recorded apart.

Match your expectations to how busy the song is and the results make sense.

What you'll need

Not much:

• A song file you already have — MP3, WAV, FLAC, or M4A, up to 5 minutes long.
• A modern browser. That's it. There is no account, no upload, and no install.

Your audio never leaves your device — all processing runs locally in your browser. The one honest asterisk: the first time you run a separation, the AI model (a few tens of megabytes) downloads once from our CDN and then stays cached, so later runs skip it. Nothing about your music is sent anywhere.

Step by step: isolating a stem

1. Open the stem splitter. You'll see a dropzone.
2. Drop in your song (MP3, WAV, FLAC, or M4A, up to 5 minutes). Separation starts automatically the moment the file lands — there's no "go" button.
3. Watch the Splitting stems panel. It shows a live percentage and a status line. It also detects your hardware and displays either GPU-accelerated (WebGPU) or CPU (WASM) so you know which path it's taking. This step takes real time — how long depends on your device and the track length — so it's not instant, especially on the first run while the model downloads.
4. When it finishes, you'll see 4 stems ready. There's a transport bar (play/pause and a seekable timeline) plus one mixer row per stem: Vocals, Drums, Bass, Other. Each row has a Solo button (headphones icon), a Mute button, a volume slider, and a Save button.
5. Solo the stem you want. Click Solo on Bass to hear only the bassline, or on Drums for just the kit. Solo and mute are live and instant — toggle them while it plays to compare.
6. Save it. Pick your export format at the bottom — WAV (default, lossless), MP3, or Opus — then click Save on that stem. It downloads in full stereo.

Want to start over with a different song? Hit Reset workspace.

Isolating a melodic instrument (the "other" stem)

Since guitars, keys, horns, and strings all share the "other" stem, this is where technique matters most.

Solo "other" and listen. On a sparse arrangement, you may already have exactly what you wanted — a lone guitar or piano line, ready to save. On a denser track, you'll hear a blend, and re-running won't separate those instruments any further, because the model can't tell apart parts that overlap in the same frequency space.

When "other" holds more than one instrument, the realistic move is to shape emphasis rather than expect perfect separation. Save the "other" stem, then load it into the free equalizer and carve away the frequency ranges you don't want. To bring out a bright guitar, for example, cut the low-mids where a piano's body sits and lift the upper mids. If your target is a low instrument, the bass booster can push it forward. Be clear-eyed about this: EQ changes which instrument dominates, but it does not fully pull two instruments apart when they occupy the same frequencies. It's emphasis, not surgery.

If your real goal is a vocal or an instrumental backing track rather than one instrument, two neighbors are better tools — the dedicated vocal remover for a two-stem vocals-versus-instrumental split, or the karaoke maker for a live mixer with key control. For the general splitting workflow, the guide to splitting a song into stems is the companion piece to this one.

Troubleshooting

My isolated guitar has other instruments bleeding into it. That's the "other" stem doing its job — it carries every non-drum, non-bass instrument at once. On a dense mix there's no clean single-guitar output to be had. Save "other" and use the equalizer to emphasize your target's frequency range, and accept that overlapping instruments won't fully separate.

I only see four stems — where's the guitar option? There isn't one, by design. The model outputs vocals, drums, bass, and other, and every melodic instrument shares "other." See stem splitting vs vocal removal if you're deciding which tool fits your goal.

The first split is slow or seems to be downloading. On the first run the tool fetches the AI model once (a one-time download), then caches it. After that, separations skip straight to processing. Speed also depends on whether you're on GPU-accelerated (WebGPU) or CPU (WASM) — the panel tells you which.

My song is longer than 5 minutes. The tool caps input at 5 minutes. Trim the section you need first with the audio cutter, then run that shorter clip through the splitter.

Frequently asked questions

Can I isolate just the guitar from a song?

Only indirectly. The AI produces four stems — vocals, drums, bass, and "other" — and guitar lives in "other" alongside keys, synths, strings, and horns. On a sparse mix where guitar is the main melodic instrument, soloing "other" gets you close. On a dense arrangement it will include other instruments, and EQ can emphasize the guitar but not fully separate it.

Which instruments separate cleanly?

Bass and drums, because each has its own dedicated stem. Solo the Bass row for a clean bassline or the Drums row for the full kit. Vocals also separate well as their own stem. Anything melodic and non-bass — guitar, piano, synths, brass, strings — is grouped into the single "other" stem, so those are harder to isolate individually.

Is the isolation perfect?

No, and no tool guarantees perfect results. AI separation is very good on typical pop and rock recordings, and weaker on dense, heavily-processed, or mono mixes. Expect clean bass and drums, a usable "other" stem on sparse songs, and some bleed on busy arrangements. It's an excellent starting point, not a flawless master multitrack.

Does my audio get uploaded anywhere?

No. Every separation runs locally in your browser, and your audio never leaves your device. The only thing downloaded is the AI model itself — a one-time fetch from our CDN, cached afterward — so your music is never sent to a server. There's no account and no upload step. To understand the technology, read how AI vocal removal works.

What's the difference between this and a vocal remover?

The stem splitter breaks a song into four parts so you can isolate instruments, while the vocal remover does a focused two-stem split into vocals and instrumental. If you specifically want acapellas, see how to isolate vocals from a song; to compare free options, the best free stem separation tools roundup covers the field.