Hear an instrumental before you upload anything
Here is a real result from the engine behind the tool. The song is Sunlight, an indie pop track we generated with TuneSplit while building the site, so the example has no copyright question. Press play on the original mix, then on the Instrumental player, and listen for three things: whether the lead voice is gone, whether the drums still hit as hard, and whether the reverb that used to trail each line is cleaner or has left a faint wash behind.
The clip is the first 60 seconds of the song and took 88 seconds of wall-clock time in one measured run. A typical 3-minute song usually takes 20 to 60 seconds. The same run also returns the vocals, drums, bass and other parts as separate stems when you choose the 4-stem option.
Two ways to get an instrumental, and when each one fits
Most pages about an instrumental maker only describe the first route. The second is often the better answer, especially when you plan to publish what you make.
| Remove vocals from a recording | Generate an original instrumental | |
|---|---|---|
| What you start with | A song file you already have | A written description |
| Where to do it | The tool on this page | The AI music maker or AI beat maker |
| What you get | The real arrangement, minus the voice | A new track written without a voice (vocal set to none) |
| Rights | Same as the source recording. You need permission to share it | Free use is non-commercial; paid plans add a commercial licence |
| Quality risk | Leftover voice, thinner reverb, small artefacts | No leftover voice, but you get what the model writes |
| Best for | Practice, private covers, remix drafts, study | Video background music, game loops, songs you plan to release |
If you need the exact backing of a particular song, use the tool here. If you need music that fits a mood and that you can publish, skip the recording and generate one. The AI music maker and the AI beat maker both run on the same song engine with the vocal option set to none, and each generation returns two versions from two different models.
What you get from a 2-stem instrumental
In 2-stem mode the model estimates the voice and subtracts it, so the output is two files. One holds everything sung. The other holds everything else. That instrumental is not a studio mix-down. It is a reconstruction, which is why a few details deserve a listen before you rely on it.
| Element in the original | In the instrumental | Note |
|---|---|---|
| Lead vocal | Removed | Goes to the vocals file |
| Harmonies, choir, ad-libs | Removed | All sung parts count as vocals |
| Drums and percussion | Kept | Hi-hats can lose a little sparkle |
| Bass | Kept | Low end is usually the cleanest part |
| Guitars, keys, strings | Kept | Lines that overlap the voice may sound slightly thinner |
| Vocal reverb and delay | Mostly removed | Tails can remain as a faint wash |
| Voice-like synths and samples | Sometimes lost | The model may file them under vocals |
If you want to hear or fix the other parts, run the 4-stem option instead: vocals, drums, bass and other. TuneSplit does not split lead from backing vocals and does not return more than four stems.
Use cases, and what each one needs from the file
The same download serves very different jobs. What matters changes with the job.
- Producers and remixers. You want to hear the arrangement, borrow a chord progression for study, or build a bootleg draft on the real drums and bass. Work from the best source file, and expect to replace anything that sounds thin.
- Video background music. A voice under your narration is the problem, so removing it fits. The catch is rights: a commercial track you do not own cannot go into a published video just because the vocal is gone.
- Cover singers. A private or practice cover needs a clean backing, a key that fits your range and no ghost of the original singer. Check the loudest chorus before you record on top.
- Rappers and writers. Instrumentals are the standard material for writing to. The tempo and groove of the original stay intact, so you can write to the real feel.
- Instrument practice. Remove the part you play so the band stays. With the 4-stem option you can also drop the bass, or keep the drums and lose the rest.
- Choir and band rehearsal. A clean backing at a slower speed, in a key the group can reach, does more than any plugin.
Make the instrumental fit: key and speed
An instrumental from a famous song is usually in a key chosen for another voice. The pitch and speed changer fixes that in your browser, with no upload. The pitch slider runs from minus 12 to plus 12 semitones, and speed runs from 50 to 150 percent without changing the pitch. Preview before you download.
- Sing or hum the highest line against the instrumental. If it strains, lower the pitch by a semitone or two and try again.
- For rap or spoken verses, pitch usually matters less than speed. Slowing a track to 90 percent often makes writing easier.
- Use the BPM and key finder if you want the tempo and key as numbers before you change anything. It runs in your browser too.
- Keep changes small. Large shifts change the character of drums and strings more than a semitone or two does.
The BPM and key finder is a separate tool. The instrumental maker on this page does not detect tempo or key itself, so run the downloaded MP3 through it if you need the numbers.
Check the instrumental in five minutes
Do this before you record, publish or build on the file. Every separation model leaves traces somewhere, and the traces appear in the same places.
- Listen to the first line. If the song starts with a bare vocal, a trace of it may open the file.
- Listen to the loudest chorus. Dense mixes leave the most ghosting there.
- Listen to the quietest bar. Reverb tails and breaths stand out when everything else drops.
- Listen to the end. A held note or vocal fade is the most common leftover.
- Switch to speakers at normal volume. A faint trace that headphones reveal often disappears in a room, but not always.
For a leftover line, a short volume dip or fade in any audio editor usually fixes it. For a section that is badly affected, use a different source file, such as a studio version in place of a live one.
Make an original instrumental from a prompt
When you want music you can publish, or you just want a track that never existed, you do not need a recording at all. In the AI music maker, set the vocal to none, describe the track, and you get two instrumental versions back. A prompt like lofi hip hop, chill, rhodes piano, vinyl crackle works well, and the AI beat maker page lists wording that steers genre, mood and instruments.
- Free use. 30 credits a day, 10 credits per generation, which is 3 generations and 6 versions. No sign-up is needed. Songs run up to 2 minutes, with an MP3 download. Free use is non-commercial.
- Paid. Pro is $9.90 a month for 450 credits, with songs up to 3 minutes, a commercial licence and WAV stems.
- Stems of your own song. A generated instrumental can be split into stems too, which is useful if you want only the drums.
Be realistic about it. The model writes the arrangement. You cannot dictate individual notes, a chord chart or an exact tempo, although a tempo in the prompt is a hint it usually follows. For total control you will still want a DAW.
Honest limits of vocal removal
Vocal removal is an estimate, not an undo button. The model never saw your original multitrack, so the instrumental is its best guess at what remains. Knowing where it fails saves time.
- Reverb tails and ghost vocals. A long reverb or delay on the voice can leave a faint, airy copy in the instrumental. It is the most common complaint, and the 5-minute check above is how to find it.
- Live recordings. Crowd noise, room sound and microphone bleed all mix into the voice, so results are weaker than on studio tracks.
- Voice-like instruments. Vocoders, talk boxes, choir pads and some synth leads share frequencies with the voice and may be partly removed or partly kept.
- Dense, loud masters. Heavily compressed mixes give the model less space between parts.
- Old mono recordings. They separate less well because the model cannot use stereo position as a clue.
- Low-bitrate sources. Compression artefacts get amplified. Start from the best file you have.
None of this makes the tool useless. Most modern studio songs come out clean enough for practice, covers and drafts. It does mean you should listen first and never assume.
Free rules, in plain numbers
| Rule | Free on TuneSplit |
|---|---|
| Files per day | 3 |
| Length per file | Up to 10 minutes |
| Formats in | MP3, WAV, M4A, FLAC, OGG, MP4 |
| Split types | 2 stems (vocals + instrumental) or 4 stems |
| Download | Allowed, every stem as MP3 |
| Model | Demucs htdemucs |
| Typical time | 20 to 60 seconds for a 3-minute song |
| Storage | Private, deleted within 24 hours |
| After the 3rd file | 30 credits per started 5 minutes |
WAV stems come with a paid plan, not the free one. Free sign-up is not needed to use the three daily files. If you need a quick count, a 3-minute song is one started 5-minute block, so one extra file costs 30 credits.
Your upload, your rights
Only upload recordings you own or are licensed to use. Making an instrumental for private practice is one thing. Posting it, putting it in a monetised video, or selling it is another, and usually needs permission from both the owner of the recording and the songwriter. Removing the vocal does not remove the copyright.
Your upload sits in private storage, is used only to produce your result, and is deleted within 24 hours. It is never used for training. TuneSplit has no voice cloning and no covers in a real singer's voice. Rights holders can use the DMCA contact on the Terms page.
Where to go next
- Vocal remover: the general page for removing vocals and keeping the music.
- Karaoke maker: the same split with lyrics and key tips for singing at home.
- Acapella extractor: the other half of the split, for remixes and DJ edits.
- Stem splitter: four stems when you want to keep drums or bass alone.
- Pitch changer: change key or speed in your browser.
- AI music maker and AI beat maker: generate an original instrumental from a prompt.
Frequently asked questions
What is an instrumental maker?
It is a tool that gives you a version of a song with no singing. Some do it by separating vocals from a recording, which is what the tool on this page does. Others generate a new track from a text prompt with the vocal switched off. Both produce an audio file you can download.
How do I make an instrumental version of a song?
Upload the song to the tool above, keep the 2-stem option, wait 20 to 60 seconds for a 3-minute track, then press play on the instrumental to check it. Download it as an MP3. Use only recordings you own or are licensed to use.
Is it free, and can I download the result?
Yes. Free means 3 files a day, up to 10 minutes each, with every stem downloadable as an MP3. No card is needed. After the third file, each started 5 minutes of audio costs 30 credits. Full-length WAV stems of your own generated songs are part of the Pro plan.
Can I make an instrumental without uploading a song?
Yes. Open the AI music maker or the AI beat maker, set the vocal to none and describe the track. You get two instrumental versions from two models. No sign-up is needed, and each browser gets 30 credits a day, which is 3 generations.
Why can I still hear a faint voice in the instrumental?
The model estimates the vocal, so reverb tails, doubled lines and voice-like synths can leave a trace. Try a lossless studio version, check the loudest chorus, and use a short fade or volume dip in an editor on any line that leaks. A live recording will always do worse.
Does it remove backing vocals and harmonies too?
Yes. Everything sung is treated as vocals in a 2-stem split, so the lead, harmonies, choirs and ad-libs all leave the instrumental. TuneSplit does not separate lead from backing vocals, so a song built on stacked harmonies can feel thinner without them.
Can I use an instrumental I made in a video or release?
Not without permission. Removing the vocal does not remove the copyright, and rights usually sit with both the recording owner and the songwriter. For music you plan to publish, generate an original instrumental instead. Free generations are non-commercial and paid plans add a commercial licence.
Can I change the key or speed of the instrumental?
Yes, with the pitch and speed changer. Pitch moves from minus 12 to plus 12 semitones and speed from 50 to 150 percent, and speed changes do not alter the pitch. It runs in your browser, so nothing is uploaded. You can preview the result and download it as a WAV.
Does the tool tell me the BPM or key of the instrumental?
Not on this page. The instrumental maker only separates audio. For tempo and key, drop the downloaded file into the BPM and key finder, which runs in your browser and gives the BPM, the key and its Camelot code. Treat both numbers as estimates and check by ear.
Which file formats work, and how long can a song be?
Upload MP3, WAV, M4A, FLAC, OGG or MP4, up to 10 minutes each on the free plan. For an MP4 video, only the audio is processed. The stems download as MP3 files, so you can load the instrumental into a phone, a video editor or a lyric video maker.
How long does it take?
A 3-minute song usually takes 20 to 60 seconds. As a real 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 the range as typical and not as a guarantee.
Is it safe to upload my music?
Files go to private storage, are used only to make your result, and are deleted within 24 hours. They are never used to train a model. Upload only audio you own or have the right to use, because you stay responsible for how the output is used.