{"id":5785,"date":"2026-11-11T20:15:45","date_gmt":"2026-11-11T17:15:45","guid":{"rendered":"https:\/\/mubert.com\/blog\/stem-splitter-free-tiers-compared"},"modified":"2026-11-11T20:16:18","modified_gmt":"2026-11-11T17:16:18","slug":"stem-splitter-free-tiers-compared","status":"publish","type":"post","link":"https:\/\/mubert.com\/blog\/stem-splitter-free-tiers-compared","title":{"rendered":"Stem Splitter Test on One Song: Six Tools, Their Free Previews and What You Can Actually Download","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"\n

A stem splitter pulls a finished song apart into separate files, usually 4 stems: vocals, drums, bass and everything else. The free tiers differ more than the sound does. A free Moises account gets 5 uploads a month, up to 5 minutes each<\/a>. LALAL.AI’s free plan lets you listen but not download<\/a>. The free tier of our own splitter<\/a> downloads full-length four-stem MP3s for 20 minutes of songs a month. This post is for producers, DJs, karaoke hosts and video editors who need a clean vocal or a drumless backing track, and it shows how to test six tools on one song and which free terms let you keep the files.<\/p>\n\n\n\n

Every stem splitter demo page sounds perfect. Every demo page also picked its own song.<\/p>\n\n\n\n

\"Stem\"Stem<\/figure>\n\n\n\n

Test every stem splitter on the same song<\/h2>\n\n\n\n

Use one song of about three minutes and run it through every tool. Three minutes fits every free tier below, including Moises’ 5-minute cap per file. Same song, same file. Any difference you hear then comes from the stem splitter, not from the material.<\/p>\n\n\n\n

Choose the song carefully. A sparse acoustic track with one voice and one guitar flatters every tool. There is almost nothing to separate. Pick something dense instead: a kick and a bass guitar fighting over the same low range, a vocal with long reverb, crashing cymbals, and a piano or guitar buried somewhere in the middle where nobody asked it to be.<\/p>\n\n\n\n

The best test song is one you mixed yourself. If you still have the multitrack project, you have the answer key. Bounce the full mix, split it, then compare each stem with the original track it came from. Researchers do roughly this. They use SDR, the signal-to-distortion ratio in decibels. Higher is closer to the real stem. Meta’s Demucs README reports 9.00 dB on the MUSDB HQ test set<\/a> for its htdemucs model. MUSDB is a research dataset of songs that come with their original stems. That figure is useful for reading papers. It does not tell you whether the hi-hat in your chorus will survive.<\/p>\n\n\n\n

Anyway.<\/p>\n\n\n\n

Set some ground rules. Upload the same file everywhere, ideally a WAV, since some tools accept fewer formats than others. Download whatever each free tier allows. Then put every vocal stem in one DAW session and switch between them blind, at the same loudness. Logos are persuasive. Ears are weak.<\/p>\n\n\n\n

Six stem splitters and their free terms, side by side<\/h2>\n\n\n\n

These six tools split a song into at least four stems: vocals, drums, bass and other. They differ in where processing runs, how many stems you get and what you can download free. The numbers below come from each stem splitter’s own help, pricing or project page as of November 2026. They change. Check again before you commit.<\/p>\n\n\n\n

None of these tools is bad. They are good at different things, and they charge for different things.<\/p>\n\n\n\n

Our stem splitter: 4 stems free, 6 on Pro<\/h3>\n\n\n\n

Mubert Stem Splitter is the one we build, and its tool page<\/a> lists the terms. It splits a song into four stems (vocals, drums, bass, the rest) or six, which adds guitar and piano. Without an account you hear a 30-second preview. A free account gets four-stem splits with full-length MP3 downloads, up to 20 minutes of songs a month. Formats: MP3, WAV, FLAC and M4A. You can also mute, solo or turn down any stem in the browser before downloading.<\/p>\n\n\n\n

Our catch: six stems need a Pro or Ultra plan, and WAV downloads start at the Creator plan. The FAQ also admits that bass and kick drum “can bleed into each other a little”. True of every stem splitter here. We are just the ones saying so on the product page.<\/p>\n\n\n\n

Moises: 5 free uploads a month, 4 stems<\/h3>\n\n\n\n

Moises’ help centre says the Free plan offers two presets<\/a>. You get 2 stems (vocals and instrumental) or 4 stems (vocals, drums, bass and other). The free cap is 5 uploads a month, and separation works on files of up to 5 minutes. Paid plans add individual instruments, and the product page<\/a> lists 27 stem types in total.<\/p>\n\n\n\n

The catch: five songs is a test, not a workflow. Long songs get chopped. Moises itself suggests cutting them into pieces under the limit as the workaround.<\/p>\n\n\n\n

LALAL.AI: ten stem types, previews free, downloads paid<\/h3>\n\n\n\n

LALAL.AI’s stem splitter lists ten separation options<\/a>. They include electric and acoustic guitar, synthesizer, strings, wind instruments and a “voice and noise” mode for dialogue. Every stem gets a short preview before you click “Split in Full”. It accepts a long list of audio and video formats and exports MP3, OGG, AAC, WAV or FLAC.<\/p>\n\n\n\n

The catch is in the pricing table<\/a>. The free Starter plan has 10 minutes of processing and no result downloads. Read that twice. Each split costs the file length multiplied by the number of stem types, so a three-minute song split into four stem types uses 12 minutes, two more than the free plan has. The cheapest paid plan, Lite, is listed at $7.50 a month.<\/p>\n\n\n\n

Demucs: free, open source, and run by you<\/h3>\n\n\n\n

Demucs<\/a> is the open-source separation model from Meta’s research team. It is MIT-licensed (a permissive open-source licence) and outputs stereo WAV files at 44.1 kHz. Its default htdemucs model splits four stems. A second model, htdemucs_6s, adds guitar and piano. Its README says plainly that the piano source “is not working great at the moment”. Refreshingly honest. Then there is htdemucs_ft, which takes four times longer and “might be a bit better”.<\/p>\n\n\n\n

The catch: you run it from the command line, on your own machine. Meta’s original repository is also no longer maintained, and only important bug fixes go to the author’s fork, which is fine for a weekend experiment and less fine for something your studio depends on every day. Spleeter, Deezer’s older open-source stem splitter, works in a similar way and does 2, 4 or 5 stems<\/a>.<\/p>\n\n\n\n

Ultimate Vocal Remover: a free desktop app with many models<\/h3>\n\n\n\n

Ultimate Vocal Remover<\/a>, or UVR, is a free, MIT-licensed desktop app. It puts a GUI (a regular point-and-click window) on top of several model families, including MDX-Net and Demucs. Windows, macOS, Linux. You can swap models and compare them on the same file without leaving the app.<\/p>\n\n\n\n

The catch is the hardware. GPU processing needs at least an Nvidia GTX 1060 with 6 GB, and 8 GB of video memory is recommended. Choosing between all those models can also eat an afternoon. Or two.<\/p>\n\n\n\n

\"Person\"Person<\/figure>\n\n\n\n

Logic Pro Stem Splitter: inside the DAW, with a one-time price<\/h3>\n\n\n\n

Logic Pro has a built-in stem splitter, and Apple added guitar and piano to it on May 28, 2025<\/a>. It has presets for acapella and instrumental versions, and the stems land straight in your project. If Apple’s other AI features in Logic are part of your setup, we covered the Logic Pro 13 assistant<\/a> separately.<\/p>\n\n\n\n

The catch: there is no free tier in the usual sense. Logic for Mac costs $199.99 for new users. The iPad version costs $4.99 a month after a one-month trial, and the splitter needs an M1 chip or later.<\/p>\n\n\n\n

What to listen for in each stem<\/h2>\n\n\n\n

A good split is mostly about leakage: how much of one instrument ends up in another instrument’s file. Listen to each stem solo, on headphones, at the same volume, and check these spots:<\/p>\n\n\n\n