Listen to a real four-stem split
The fastest way to judge a stem splitter is to hear one. The players below contain a real split made by the same engine as the tool on this page. The source is Sunlight, an indie pop song we generated with TuneSplit, so you can use it without any rights question. You get the original mix, then vocals, drums, bass, other, and an instrumental made by combining everything except the voice.
This split covers the first 60 seconds of the song and took 88 seconds of wall-clock time, from upload to finished files. That is one measured run, not a guarantee. For a typical 3-minute song, the usual range is 20 to 60 seconds. Play the drums and bass stems loudly and listen for two things: how cleanly the kick and snare come through without the synths, and whether the bass holds together without picking up vocal sibilance.
The other stem is the catch-all. It holds whatever the model does not call vocals, drums or bass: synths, guitars, piano, strings, pads, effects. In this example it is mostly warm synths and claps. In a band recording it would contain the guitars and keys.
Stem splitter facts at a glance
The key facts in one table. Every line describes how the splitter on this page works today.
| Fact | TuneSplit stem splitter |
|---|---|
| Free download | Yes, every day: 3 files, each up to 10 minutes, every stem as MP3 |
| Sign-up | Not needed. No registration, no email, no card |
| File deletion | Uploads and stems are deleted within 24 hours |
| Training | Your audio is never used to train any model |
| Stems | 4 (vocals, drums, bass, other) or 2 (vocals and music) |
| Model | Demucs htdemucs, open-source source separation |
| Speed | Usually 20 to 60 seconds for a 3-minute song |
| Install | Nothing to install; works in a phone or desktop browser |
What each of the four stems contains
Four stems is the most common layout in music separation because it matches how most songs are built. Here is what ends up in each file.
| Stem | What you usually find in it | Typical use |
|---|---|---|
| Vocals | Lead vocal, backing vocals, harmonies, spoken parts | Acapellas, remix material, vocal practice |
| Drums | Kick, snare, hi-hats, cymbals, percussion, drum machines | Sampling breaks, drum covers, tempo work |
| Bass | Bass guitar, synth bass, 808s, low sub layers | Learning bass lines, remix foundations |
| Other | Guitars, keys, synths, strings, pads, effects | Chord study, mashups, backing for singers |
If you only want to take the voice out, the two-stem option is simpler: one vocals file and one music file. The vocal remover page is written around that case, and the vocal isolator page around the opposite one, keeping only the voice.
How the separation works, and what to expect
TuneSplit uses Demucs, in its htdemucs version. Demucs is a neural network trained on multitrack recordings where each instrument exists as its own file. It learns what a voice, a drum kit and a bass line look like inside a mixed signal, then estimates each one in a song it has never seen. The htdemucs model looks at both the waveform and the spectrogram, which tends to help with sharp attacks like snares as well as sustained sounds like pads.
Estimating is the key word. A stem splitter reconstructs what it thinks each layer was, so the stems are very good but not identical to the original studio tracks. Mixes that are heavily compressed, saturated, or loud with wide reverb are harder than clean modern pop or hip hop. You may hear a trace of the hi-hats inside the vocals stem, or a faint echo of the singer in the other stem. That is normal in this category, and it is the reason we let you listen before you download.
A few things work well in practice. Start from a lossless file if you have one. Avoid re-encoded MP3s. For percussive loops, check the drums stem on its own before you sample it. For bass, listen on headphones or monitors that can actually reproduce low frequencies, because a laptop speaker will hide problems.
What you can do with separate stems
- DJ sets and live mixes. Drop a vocals stem over another track's instrumental, or loop the drums and bass for a longer transition.
- Remixes and mashups of music you have the rights to. Rebuild the arrangement, change the drums, keep the vocal.
- Learning songs. Mute the bass or the guitar part and play along, or isolate the bass to learn the line by ear.
- Drum and bass practice. Play with the full band minus your own instrument.
- Sampling your own recordings. Pull a break or a vocal line out of an old demo you no longer have the multitrack for.
- Video, podcast and stream work. Remove music under dialogue, or lower the music while keeping the vocal sharp.
- Mixing and mastering checks. Hear how the bass behaves against the kick when they are not masking each other.
Each of these needs the original recording to be yours or licensed for the use you have in mind. Splitting a stem is a technical step, not a licence.
Free limits, stated clearly
The free plan has plain numbers. These are the numbers that matter before you upload.
| Rule | TuneSplit free |
|---|---|
| Files per day | 3 |
| Length per file | Up to 10 minutes |
| Formats in | MP3, WAV, M4A, FLAC, OGG, MP4 |
| Stem layouts | 2 stems or 4 stems |
| Download | Every stem as MP3, free |
| Model | Demucs htdemucs |
| Typical time | 20 to 60 seconds for a 3-minute song |
| Storage | Private, deleted within 24 hours |
| Beyond 3 files | 30 credits per started 5 minutes |
Beyond three files in a day, each started five minutes of audio costs 30 credits. Credit packs are 200 credits for $4.90 and 600 credits for $12.90, valid for 12 months. At those prices a 3-minute song costs roughly 65 to 74 cents. The free daily files do not need credits at all.
TuneSplit vs StemSplit, LALAL.AI and Moises
Stem splitting is a crowded category, and the tools differ more in their rules than in their models. This table compares what we could confirm in October 2026. Moises prices come from third-party sources, so confirm them on its site.
| Tool | Free option | How you pay (as of Oct 2026) | Notes |
|---|---|---|---|
| TuneSplit | 3 files a day, up to 10 min, MP3 download | 30 credits per started 5 min after the free files | 2 or 4 stems, private storage, deleted within 24 hours |
| StemSplit | Free trial; details on its page | Credit packs of $2, $5, $12 and $30, about $0.20 down to $0.10 a minute, credits do not expire | Drag and drop, lists a 100 MB file limit |
| LALAL.AI | 10 minutes, preview only, no download | Lite $9.99 and Pro $19.99 a month; 750 minutes for $50, 3,000 for $190 | Wide range of stem types |
| Moises | About 5 uploads a month (unverified) | Premium $3.99 to $5.99, Pro $9.99 to $29.99 a month (unverified) | Musician apps and practice features |
Be clear-eyed about the trade-offs. If you split many hours of audio a month, a credit pack or a subscription from any of these services can be cheaper per minute than occasional use of ours. LALAL.AI separates more instrument types than the standard four. Moises is built for practicing musicians and has features around the stems. StemSplit's pay-per-minute pricing is easy to understand.
TuneSplit's position is smaller: a free daily allowance that you can actually download, a limit of 10 minutes per file, two layouts, and clear privacy terms. That is a good fit for the occasional split, which is what most people searching for a stem splitter want.
TuneSplit vs vocalremover.org, UVR, Fadr and BandLab
When people ask an AI assistant how to split a song into stems for free, these four names come up most. The table compares them on where they run, sign-up, and whether the free stems can be downloaded.
| Tool | Where it runs | Sign-up | Free download | Stems | Paid (as of Oct 2026) |
|---|---|---|---|---|---|
| TuneSplit | Any browser, phone or computer | Not needed | Yes, MP3, 3 files a day up to 10 min each | 4 or 2 | Credits after the free files (credit packs) |
| vocalremover.org | Browser | Not needed, per reviews (not verified) | About 1 file a day up to 10 min, per reviews (not verified) | Vocals + music | Not verified |
| Ultimate Vocal Remover (UVR) | Desktop app you install (Windows 10+, macOS) | Not needed | Free and open source, runs on your own computer | Depends on the model you download | Free |
| Fadr | Browser | Not verified | MP3 on the free plan, per reviews; WAV needs Plus | Vocals, drums, bass, melodies, per reviews | Plus about $10 a month (third-party reviews) |
| BandLab Splitter | Browser and BandLab app | BandLab registration, per reviews (not verified) | Free; a cap of about 2 splits a day is reported (not verified) | 2 or 4 (vocals, drums, bass, other) | Membership about $8.25 a month billed yearly (third-party) |
Each has a real strength. UVR is free, open source and offline, but you install a desktop app, download models and wait on your own hardware. BandLab and Fadr put splitting inside larger music apps, and vocalremover.org is a long-running simple tool.
TuneSplit keeps it to one page: no install, no sign-up, all four stems as free MP3 downloads every day, and deletion within 24 hours, from any browser including a phone.
Checked on October 18, 2026. UVR facts come from its official GitHub README. For the others we used their public pages and, where a site blocked automated checks, third-party reviews; anything we could not confirm is marked not verified. Prices change, so check each site before you decide.
Choosing between 2 stems and 4 stems
Use 2 stems when the job is binary. Karaoke, a vocal-free backing track, or a cappella extraction all need exactly two files, and the model has less to guess about when it only separates the voice from everything else.
Use 4 stems when you need to work on the instruments themselves: reinforcing the kick, replacing the bass, muting the guitar, or building a DJ blend. It costs the same one file from your daily allowance, so if you are not sure, choose 4 stems and combine files afterwards.
A related trick is to build your own custom backing track. Load the four stems into any DAW, mute the one you want to play yourself, and bounce the rest. This is more flexible than an instrumental file, because you can set the volume of each layer to taste.
Privacy and rights
Only upload audio that you own or have the right to use. Stems you make from your own songs, licensed material, or tracks with clear permission are fine. Stems of commercial releases are for private study unless you have a licence for the use you plan.
Your upload goes to private storage, is used only to make your stems, and is deleted within 24 hours. It is never used to train a model. TuneSplit does not offer voice cloning or covers in a real singer's voice, and lyrics in generated songs are moderated. The Terms page has a DMCA contact for rights holders.
Frequently asked questions
What is a stem splitter?
A stem splitter is a tool that separates a finished song into its component layers, called stems. TuneSplit gives you four: vocals, drums, bass and other. It does this with a neural network, Demucs htdemucs, that estimates each layer from the stereo mix, so you do not need the original multitrack session.
Is the stem splitter free?
Yes, within limits. You get 3 files a day, up to 10 minutes each, and every stem downloads as MP3. That is a free download, not just a preview. Past the third file in a day, each started 5 minutes of audio costs 30 credits.
How many stems do I get?
You can choose 2 stems, which gives vocals and music, or 4 stems, which gives vocals, drums, bass and other. Both layouts count as one file against your daily allowance. The other stem collects everything that is not vocals, drums or bass, such as guitars, keys and synths.
How long does stem splitting take?
A 3-minute song usually takes 20 to 60 seconds. For reference, the 60-second four-stem example above took 88 seconds in one measured run. Time grows with file length, and it can go up when the service is busy, so treat these as typical numbers, not promises.
What file formats can I upload and download?
You can upload MP3, WAV, M4A, FLAC, OGG and MP4 files. Stems download as MP3. If you need uncompressed output for professional work, start from a lossless source file for best results and keep in mind that the downloaded stems are MP3.
Will the stems be as clean as studio multitracks?
No. Stems from any AI splitter are estimates. On clean modern mixes they can be very usable, and on dense or heavily processed mixes you may hear bleed, such as a hint of hi-hat in the vocal. Listen to each stem before you build on it.
Does it work on a phone?
Yes. The splitter runs in a mobile browser, with nothing to install. Tap the upload box, pick a file on your iPhone or Android phone, and wait for the stems to appear. The processing happens on our servers, so a fast connection helps with large files.
Are my files private?
Uploads go to private storage, are used only to make your stems, and are deleted within 24 hours. They are never used for training. Please upload only audio that you own or have the rights to, because you remain responsible for how you use the stems.
Can I use the stems in my own music or videos?
It depends on who owns the recording and the composition, not on the tool. If you made the song, or you hold a licence that allows remixing and sampling, you are fine. For other people's music, splitting does not give you the right to publish the result.
How does TuneSplit compare with StemSplit, LALAL.AI and Moises?
The core idea is the same, so the differences are rules and price. TuneSplit offers 3 downloadable files a day, with 2 or 4 stems. LALAL.AI's free tier is preview only, StemSplit sells credit packs at about $0.10 to $0.20 a minute, and Moises focuses on musician features.