Hear the result before you try your own file
If you want to know what you will get, play a real split first. The players below hold one song separated by the same Demucs htdemucs model that powers the tool above. The song is Sunlight, an indie pop track we generated with TuneSplit while building the site, so there is no copyright question about the example. Play the original, then switch to the Instrumental player and notice that the lead voice is gone while the beat and the synths stay in place.
The clip covers the first 60 seconds and took 88 seconds of wall-clock time in one measured run. That is one data point, not a promise. For a typical 3-minute song the usual range is 20 to 60 seconds, and busy periods can add waiting time.
On your phone: iPhone and Android
Most people who search for this are holding a phone, and a phone is enough. The tool runs in the mobile browser, so there is no app to install and nothing to pay for before you can listen to the result. The separation happens on our servers, not on your device, so an older phone is not a problem.
- Open this page in Safari on iPhone or in Chrome on Android, and scroll to the upload box at the top.
- Tap the box. On iPhone you can choose Photo Library, Take Photo or Choose File. Pick Choose File to browse the Files app, which includes iCloud Drive and On My iPhone. On Android the system file picker lists Downloads, Drive and recent audio.
- Choose the song. A song bought or ripped to your device works; a song that only streams inside a music app does not exist as a file you can pick.
- Wait for the upload bar to finish. Use Wi-Fi if you can, because a 5 MB MP3 is quick on mobile data but a 60 MB FLAC is not.
- Play the Instrumental player, then tap Download. On iPhone the MP3 lands in the Downloads folder in the Files app. On Android it appears in the Downloads folder and in your music apps.
A voice memo or a video with music in the background is a normal input too. If the phone locks during the upload, return to the page and retry.
On a computer: Windows, Mac and Linux
On a desktop the steps are the same, and drag and drop saves a click. Drag the audio file from your folder onto the upload box, or click the box and use the file picker. Any current Chrome, Safari, Edge or Firefox works.
A computer helps most with what comes next. Import the instrumental into a DAW such as GarageBand, Audacity, Reaper or Ableton to record vocals over it, change the key, or trim an intro. A browser cannot do that for you, but the MP3 opens in every editor.
Which files work and what you get back
The tool accepts the common audio formats and returns MP3 on the free plan. Feeding it the best-quality file you have gives the model more to work with.
| Input format | Works? | Notes |
|---|---|---|
| MP3 | Yes | The most common input. A 192 kbps or higher file is a good starting point. |
| WAV | Yes | Uncompressed. Large, so uploads take longer on a phone. |
| FLAC | Yes | Lossless and smaller than WAV. The best source for clean results. |
| M4A / AAC | Yes | What iPhone voice memos and many store downloads use. |
| OGG | Yes | Common with open-source players and some games. |
| MP4 video | Yes | The audio track is processed. Handy for music videos you own. |
On the free plan every stem downloads as an MP3. Full-length WAV stems are offered for songs generated on TuneSplit.
Check you have the right to use the song
Taking the vocal out of a recording does not change who owns it. Before you upload, ask what you plan to do with the result. The rules below are general guidance, not legal advice, and local law varies.
| Your plan | Typically fine? | What to check |
|---|---|---|
| Practise singing or an instrument at home | Generally yes | Keep it private; do not post or sell it |
| Your own song, demo or session | Yes | You hold the rights to the recording and the composition |
| Post a cover to YouTube or TikTok | It depends | You may need a licence for the composition and the backing track |
| Sell or stream a release using someone else's instrumental | No, not without a licence | Get written permission from the rights holders |
On TuneSplit, only upload audio you own or are licensed to use. The tool asks you to confirm this before a file is processed. Free split files are for personal use. If you want an instrumental you can publish with no questions, generate an original song on the site, which has no rights problem, and split that one. Paid plans add a commercial licence. The pricing page has the current plans and prices.
Troubleshooting: when a little vocal is still there
AI separation is an estimate, so a clean result is common but not guaranteed. The most frequent problems have predictable causes. Listen to the whole instrumental before downloading, headphones on.
| What you hear | Why it happens | What to try |
|---|---|---|
| A faint ghost of the voice | The model leaves a little energy where a vocal overlaps synths or guitars | Try a studio or remastered version of the track |
| Echo or reverb tails after each line | Reverb is a long wash that blends into the music | Fade or trim at phrase ends, or use a gate in an editor |
| Backing vocals and harmonies remain | Some stacked parts sound like instruments to the model | Use the 4-stem option and compare, or edit the extra parts out by hand |
| Choppy or underwater sound | Low-bitrate source | Start from a lossless or higher bitrate file |
| Crowd noise on a live recording | Applause and room sound are not music or voice | Use a studio version if one exists |
| Mono file sounds odd | There is no stereo information to help | See the mono section below |
Backing vocals are harder than the lead, because stacked harmonies blend into the pads beneath them. Mono files have no left-right difference, so centre-cancel methods cannot work on them, while a model-based split can still run. Distorted guitar walls, vocoders and choirs are the hardest material for any tool. A short fade at the ends of lines often hides what is left.
What about Audacity and the phase-cancellation trick?
You will find tutorials that remove vocals in Audacity without any AI. They work on a real, if narrow, principle, and it helps to understand why they usually disappoint.
In many stereo recordings the lead vocal is mixed to the exact centre, meaning it is identical in the left and right channels. If you invert one channel and mix the two together, anything identical in both cancels out. The vocal goes away. In Audacity this is either the Vocal Reduction and Isolation effect or the manual route of splitting the stereo track, inverting one side, and mixing down to mono.
The trouble is that the vocal is not the only thing in the centre. Kick, snare and bass are normally centred too, so they vanish with it, and the result sounds thin and hollow. Stereo reverb on the voice does not cancel, because the reverb differs between channels, so the singer's echo stays behind. Any vocal that was panned, doubled or recorded in stereo stays too. And the output is mono, losing the stereo width of the track.
| Phase cancellation (Audacity) | AI separation (this tool) | |
|---|---|---|
| How it works | Subtracts what is identical in both channels | A neural network estimates the voice and the music |
| Bass and kick | Often cancelled with the vocal | Normally kept in the instrumental |
| Reverb tail | Stays, since it differs by channel | Partly removed, partly left |
| Stereo width | Output is mono | Instrumental stays stereo |
| Mono source | Does not work | Works, usually with less clarity |
To be fair to the old method: it has no upload and no limit, and on an older recording with a perfectly centred lead it can be passable. For modern stereo mixes, the AI approach is a clear step up. A sensible workflow is to use the AI file as your instrumental, then finish it in Audacity or any editor with a fade or a gate.
Remove vocals from a song: facts at a glance
| Item | TuneSplit |
|---|---|
| Price (free tier) | Free: 3 files a day, up to 10 minutes each |
| Download format | MP3 on the free plan |
| Sign-up | Not needed for the free files |
| File deletion | Uploads are deleted within 24 hours, never used for training |
| Model | Demucs htdemucs, 2-stem or 4-stem split |
| Speed | Usually 20 to 60 seconds for a 3-minute song |
| Limits | 10 minutes per file; after the free files, 30 credits per started 5 minutes |
| Use | Free files are for personal use; paid adds a commercial licence |
What to do with the instrumental
- Sing along. Play the MP3 from any phone or speaker, with the lyrics on a second screen.
- Practise an instrument. Play the missing parts yourself, or slow the track down with the tempo changer while you learn a tricky section.
- Change the key. Use the pitch changer to move the instrumental up or down a few semitones so it fits your voice.
- Check the tempo and key. The BPM and key finder tells you the BPM and the Camelot code if you plan to blend tracks.
- Record over it. Drop the MP3 into a DAW and record a new vocal or a cover for private use or with the right licences.
Related tools on TuneSplit
- Vocal remover: the full tool page, with free rules and a comparison of other services.
- Karaoke maker: turn the instrumental into a sing-along session.
- Vocal isolator: keep the voice and drop the music instead.
- Stem splitter: separate vocals, drums, bass and other.
- Pitch and speed changer: set the instrumental to your key and tempo.
- BPM and key finder: find the tempo and key before you mix.
Frequently asked questions
How do I remove vocals from a song for free?
Upload the song to the tool at the top of this page, wait 20 to 60 seconds for a 3-minute track, and download the instrumental as an MP3. The free plan covers 3 files a day, up to 10 minutes each, with no sign-up. Uploads are deleted within 24 hours.
Can I remove vocals from a song on my iPhone?
Yes. Open this page in Safari, tap the upload box, choose Choose File to browse the Files app, and select the song. When it finishes, tap Download on the Instrumental player. The MP3 saves to the Downloads folder in Files. Nothing needs to be installed, and processing happens on our servers.
Can I remove vocals on Android?
Yes. Open the page in Chrome, tap the upload box, and pick the song from Downloads, Drive or your audio files. After the split, tap Download on the Instrumental player and the MP3 lands in your Downloads folder. A song that only streams inside a music app is not a file, so it cannot be picked.
Can I take the vocals out of a YouTube or Spotify song?
Not directly. The tool needs an audio file you have the rights to use, and a streaming-only track is not one you can upload. If you own a copy of the song, such as a purchase or your own recording, you can upload that. Only use audio you own or are licensed to use.
Why can I still hear the singer a little?
AI separation estimates the voice, so a faint trace can remain, especially where the vocal overlaps guitars or synths, or in reverb tails. Try a studio or lossless version of the song, listen with headphones before downloading, and trim or fade the ends of phrases in an editor if needed.
Does it work on live recordings and mono files?
It does, with caveats. Live recordings add crowd noise and room echo that blur the line between voice and music, so expect more leftover. A mono file has no left-right difference, so the old cancellation trick fails, while a model-based split can still run, usually with a little less clarity than stereo.
Is removing vocals from a song legal?
Removing a vocal for private practice or karaoke is generally low risk, but publishing, streaming or selling a track made from someone else's recording usually needs permission. This is general information, not legal advice. TuneSplit asks you to confirm you own the audio or have the rights before a file is processed.
Why is Audacity's vocal removal so much worse?
It subtracts whatever is identical in both channels. The vocal is often there, but so are kick, snare and bass, so they disappear too, while stereo reverb stays behind and the output is mono. AI separation looks at what each sound is, so it keeps the rhythm section and the stereo image far better.
How long does it take?
A 3-minute song usually takes 20 to 60 seconds, plus the upload time. As a reference, the 60-second example on this page took 88 seconds in one measured run. Longer files take longer, and busy periods can add waiting time, so treat these numbers as typical rather than guaranteed.
Do I need to sign up?
No. The free files need no sign-up, and the daily limit is 3 files of up to 10 minutes each. After the free files, each started 5 minutes costs 30 credits. Paid plans are described on the pricing page, and they add a commercial licence as well as more capacity.
Can I keep the vocals as well as the instrumental?
Yes. A 2-stem split returns both files from a single upload: the Instrumental and the Vocals (acapella). Each has its own player and its own Download button, and both download as MP3 on the free plan. That means you only use one of your 3 daily files for both halves.
Is it safe to upload my song?
Uploads go to private storage, are used only to produce your result, and are deleted within 24 hours. They are never used for training. Even so, upload only what you own or have the rights to, and avoid private material you would not want stored for a day.