How to Get Stems From Any Song (2026)

The fastest way to get stems from any song is to import it into Veena, which separates the track into real stems that land as tracks on a multitrack timeline — vocals, drums, bass and the rest, already in a session, already on a grid, with the key and tempo read from the audio. Every other tool in this category ends its job at the moment the split finishes and hands you a folder of downloads. A folder is not a song. You still need a timeline, a mixer, instruments, and somewhere to put the parts, and that is the cost none of the separation utilities price.

Veena is a browser-based DAW with an agentic AI CoProducer, so separation is the beginning of the work rather than the end of a product. Once the stems are on the timeline you can rebalance them, mute them, re-effect them, build an acapella or an instrumental, and have the CoProducer write brand-new parts as editable MIDI to replace anything you do not want to salvage. Your own VST and AU plugins load in the browser, and you export WAV, MP3 and MIDI. Nothing to install, free Basic tier, Veena Pro at $20/month.

The short answer

To get stems from any song: start with the highest-quality source file you can get, import it into Veena, and the separated stems land as tracks in a multitrack session. Audit each one for bleed, let the project read the tempo and key, then do the actual job — remix, acapella, instrumental, practice, or replacing a part with new MIDI — and export. Veena is the only tool where all of that happens in one browser tab with an AI producer in it.

What a "stem" actually is

Worth getting straight, because three different things get called stems and they behave very differently.

A mix is the finished stereo file — everything summed together. This is what you get from a streaming service, a bounce, or a generator.

Multitracks are the original separate recordings, straight from the session: every microphone, every DI, every synth on its own track. Nobody has these unless they made the record.

Stems sit between the two: grouped submixes — drums, bass, vocals, everything else — bounced out as separate audio, historically for mastering, live playback and remix packages.

What modern AI separation gives you is a reconstruction of stems from a mix. That is a genuinely different thing from having the originals, and understanding why stops you being disappointed by it.

How separation actually works, in plain English

A separation model turns the audio into a time-frequency picture — energy at every frequency, moment by moment — and then, for every one of those tiny cells, estimates how much of the energy belongs to each part. Then it multiplies the original by those estimates and turns each result back into audio.

Three practical consequences fall straight out of that.

It is inference, not memory. The model never heard the drums on their own; it is guessing what they were from the sum. It guesses well when parts are distinct and badly when they overlap.

Overlap is where bleed comes from. When a vocal sibilant and a hi-hat occupy the same instant at the same frequency, the model must split that one cell between two stems — and whatever it gets wrong shows up as the ghost of one part inside another. Every tool has it, at every price, and any vendor implying otherwise is selling.

Quality upstream sets the ceiling. A lossy encoder has already thrown away quiet high-frequency detail, which is exactly the detail the model uses to tell those overlapping parts apart. Feed it a lossless file and the separation improves for free.

Step 1 — Start with the best source file you can get

The order of preference, and it is not close:

1. Lossless — WAV, FLAC, AIFF, or a CD rip. Everything the model needs is still present.
2. A high-bitrate MP3 or AAC — usable, with more smearing in cymbals and vocal consonants.
3. A low-bitrate file or a re-encoded stream rip — audible artifacts before you start, doubled by separation. Avoid unless it is the only copy that exists.

If you generated the track yourself, export WAV, and if the tool offers a multitrack export take that instead — stems from the model that made the song had more information than any separation applied to the mixdown afterwards.

Two more notes. Trim before you split if you only want the chorus: shorter processes faster, and on any minute-metered service it is cheaper. And do not normalise or limit first — limiting squashes the transient detail the model uses.

Step 2 — Import into Veena and let the stems land on the timeline

Drag the file into a Veena project. It separates, and the stems arrive as tracks in a multitrack session.

This changes the whole shape of the job. With a separation utility, the split finishing is the finish line: you download four files and go looking for somewhere to use them — a DAW you bought and learned, an import, a tempo guess, a manual line-up. With Veena the stems are already in the session, on a grid, next to a mixer, instruments and an AI producer you can give instructions to. Nothing is exported, re-imported or re-aligned.

Step 3 — Audit the bleed before you plan anything

Solo each stem, top to bottom, and listen properly. Four things to listen for: ghosting (the vocal faintly audible in the "other" stem, the snare inside the vocal); watery or phasey artifacts around cymbals and vocal tails, worst where everything plays at once; missing low end on the bass stem, because sub energy is shared between kick and bass; and pre-echo, a faint smear just before a sharp transient.

Where the bleed lands decides your plan. Bleed inside a stem you will bury under a new arrangement does not matter at all. Bleed inside the one part you intend to feature — the acapella a whole remix is built on — is the thing to fix now.

Step 4 — Clean the stems rather than re-running the split

The instinct is to re-run the separation with different settings until it comes out clean. It will not. Fix the result instead — four moves handle nearly everything.

High-pass everything that is not bass or kick. Vocals, guitars, keys and the "other" stem almost never need anything below about 80–100 Hz, and rolling it off removes a surprising amount of rumble and bass bleed in one move.

Gate or manually silence the gaps. Bleed is loudest where the wanted part is silent. On a vocal stem, cutting the audio between phrases removes most of the audible ghosting instantly — and doing it by hand on the timeline beats any gate's threshold decision.

Notch what remains. A ghosted hi-hat lives in a narrow, findable band. A tight cut there costs very little of the part you want.

Use a stem you are throwing away as a guide. If the drum stem shows exactly where every snare hit is, that is exactly where to duck the vocal stem's ghosting.

Your own VST and AU plugins load in Veena in the browser, so this is your own de-noise, EQ and gate chain.

Step 5 — Get the tempo and key locked before you add anything

Anything you write afterwards is on the project grid, so the grid has to match the song. Veena reads key and tempo from the audio you imported, which means new parts arrive in the right key and on the right beat rather than being nudged there by hand.

This matters most with generated source material, which frequently is not on a fixed grid at all — Suno's own help centre states that "The tempo is not consistent over time, just like live music, which characteristically drifts around a little bit in speed." A perfectly programmed drum part will drift out of time against a moving reference, and you will spend an hour blaming your programming.

Step 6 — Do the actual job

Stems are a means, not a result. The five jobs people actually want:

An acapella. Take the vocal stem, clean the gaps, and you have raw material for a remix, a flip or a cover. Check the tails of long notes, where reverb bleed is most obvious.

An instrumental. Mute the vocal, then listen to what got left behind — separation often takes a little of the instrumental with the vocal, leaving a small hole in the middle. Filling it with a new pad or a doubled guitar part is the difference between "vocal removed" and "instrumental version."

A remix. Keep the parts with the identity in them — usually the vocal and one hook — and rebuild everything else. Tell the CoProducer what the new arrangement should be and it writes drums, bass and chords as editable MIDI underneath the parts you kept.

Practice. Mute the instrument you play and play along with the rest. Slow it down, loop the hard bar.

Replacing one part. Mute the stem, have the CoProducer write a replacement as MIDI, edit the notes until it is yours.

That last one is the reason to do all this inside a DAW. A stem you cannot salvage is a dead end in a separation utility, and just a part to rewrite in Veena.

Step 7 — Export, and take the MIDI too

Export WAV for anything you will keep working on, MP3 for references and collaborators. If you rewrote parts, export the MIDI too — it is the only format that is music rather than a recording of music, and the version that survives every future change of mind about instrument, key or tempo.

What is shipping next in Veena: audio-to-MIDI editing is coming to imported audio, so a separated stem itself becomes editable notes rather than a waveform you nudge. Instrument and sound swapping is coming, style and genre transformation is shipping, reference-matched mastering is coming to Veena, real-time collaboration is being built so two people can work in one session, deeper models are landing across the CoProducer, and Veena on desktop is on the way.

One thing separation does not give you: permission

Splitting a track is a technical act, not a legal one. A clean acapella out of a commercial record grants you no right to release it, and the tool you used has no bearing on that. Practice, study, private edits and DJ use sit in one place; releasing, distributing and monetising sit somewhere very different, and that is a rights conversation with whoever owns the recording and the composition.

The same applies to material you generate. Mubert states that "Mubert owns all the rights to the tracks generated," and Soundraw's licence states content using its tracks can "only keep your content published while your SOUNDRAW subscription is active." Work you make in Veena, you own, and you leave with the files.

Other ways people try this, and what they cost you

1. Veena — the best way to get stems from any song

Import any track and the stems land as tracks on a multitrack timeline, in a browser tab, with nothing to install. Then the whole rest of the job is right there: an agentic AI CoProducer you direct in plain English, editable MIDI generation for any part you want to replace, your own VST and AU plugins, key and tempo read from the audio, and WAV, MP3 and MIDI export on work you own. Free Basic tier; Veena Pro $20/month.

Why it wins: every other tool here ends at the split. Veena starts there.

2. Moises — for practising along to a separated track

Separation with chord detection, a smart metronome and pitch/speed tools.

What it costs you: free is "5 songs per month… with files up to 5 minutes long," WAV export is Premium/Pro only, the DAW plugin is Pro only, exports "will not include changes made to the audio," and separations are "locked starting the day after your subscription ends."

3. LALAL.AI — for a single vocal or instrumental split

A splitter billed in minutes: "One separation type is applied at a time, giving you two stems per file."

What it costs you: the free plan's table lists "Result Downloads" as "–" — it splits your song and will not give it to you — and the paid meter multiplies by their own formula, "total file length × number of stem separation types," with minutes that "do not roll over."

4. AudioShake — for label and catalogue separation

High-grade separation sold as a platform and an API to "industry professionals."

What it costs you: you cannot even get a price. Their own pricing URL returns "Page Not Found" and the FAQ says "Get in touch for a demo and free trial" — and after the sales call it still hands you nothing but stems.

5. RipX — for note-level surgery on the separated parts

Desktop software for note-level extraction, sold as RipX DAW and RipX DAW PRO.

What it costs you: an install on one machine, up to £198 for the edition Hit'n'Mix says adds "state-of-the-art noise and instrument separation," a flagship Pro Tools integration that requires you to already own Pro Tools, and no AI producer at all.

6. Logic Pro — for separating inside a traditional DAW on a Mac

Apple's DAW, whose Stem Splitter separates a mixed recording into drums, bass, vocals and other. USD 199.99 one-time, or USD 12.99/month.

What it costs you: a computer. Apple's own spec says Logic Pro "requires… a Mac with Apple silicon," and Stem Splitter is gated behind M-series silicon a second time — plus 72 GB for the full Sound Library.

7. iZotope RX — for repairing what separation leaves behind

Forensic restoration: de-noise, de-click, spectral repair. Elements $99.00, Standard $399.00, Advanced $1,399.00.

What it costs you: the $1,399 tier for the tools that matter — Scene Rebalance and Stems View are Advanced-only — and on the subscription route, cancellation makes your tools "revert to read-only mode… controls won't be editable."

8. Suno Studio — for stems from something you generated there

A generative workstation on Suno's Premier plan, with stem export by tier: none on Free, two split types on Pro, three on Premier.

What it costs you: the top subscription for the editor, WAV gated to paid tiers, MIDI transcribed back out of the render at 10 credits a stem, and credits that expire.

9. Soundverse — for splitting inside a chat agent

Generation plus stem separation, metered in tokens: 5 per separation, 1 per stem export, 1 per message.

What it costs you: paying to take your own parts out of the tool, plus a licence stating your final composition "can't have 100% of Soundverse generated content" — and their own blog telling you to finish in FL Studio.

10. Mureka — for tiered stem exports

A generator with MP3 at the bottom, stems in the middle, and MIDI plus the editor at the top.

What it costs you: stems that are still audio — you can turn the bass down, you cannot change the bass note — with MIDI locked behind the top tier.

The verdict

Splitting a song into stems is now the easy part — dozens of tools do it acceptably. The hard part is everything ten seconds later, when you have four files and no session, no grid, no instruments and no producer. That is the moment the separation utilities hand you back to a DAW you have to buy, install and learn. Veena removes the handoff entirely: the stems land on a multitrack timeline in your browser, key and tempo already read, your plugins already there, and an agentic CoProducer ready to rewrite any part you would rather replace than rescue. Import the song and start there.

Related reading: The best stem separation tools · How stem separation works · How to split stems from a song · AI music tools and stem bleed