- Blog
- Guide
How to Remove Vocals From a Song: 3 Easy Ways
Make a karaoke track or an instrumental from any song: with AI, with free tools or in an audio editor. Real results compared, so you can hear the difference.
· 7 min read
A karaoke night, a cover for YouTube, a backing track to practice on, or a remix: you need the song without its singer. For years the only way was an old stereo trick that sounded hollow. Today AI separates the voice from the music cleanly, in about a minute. Here are three ways to remove vocals from a song, with real results so you can hear the difference.
The short answer
- Best result, any device: an AI vocal remover. Drop the song, listen to the preview, download the instrumental and the vocals.
- Free and instant: center channel cancellation, built into many free tools. It removes the voice only on some songs, and takes the bass with it.
- In an audio editor: Audacity with its free AI plugin, or the stem splitter of a recent DAW. More control, more time.
Hear the difference
We made a short test song, then removed its vocals with the AI and with the free trick. Press play on each line:
Original song · voice and music
AI: instrumental · the voice is gone, the beat stays
AI: vocals only · the voice alone, an acapella
Free trick: center cancellation · voice gone, but the kick and bass too
Both remove the voice. But look at the red waveform: the regular peaks of the kick drum are gone, and so is the bass. We measured it on the 20 seconds:
| Measured | AI vocal remover | Center cancellation |
|---|---|---|
| Voice left in the instrumental | 6% | 0.1% |
| Kick drum and bass kept | 100% | 0% |
| Result | Stereo, full song | Mono, no low end |
This test is the best case for the free trick: our voice is dry and exactly in the center. On a real song, reverb, echoes and doubled vocals are spread across the stereo field, so cancellation leaves a ghost of the voice too. The AI, on the other hand, found the voice back with 99% correlation with the original voice track we mixed in.
The three ways side by side
| Way | Quality | Cost | Time | Effort |
|---|---|---|---|---|
| AI vocal remover | Very good | Free preview, €1.99 per song | About a minute | Drop the song, listen, download |
| Free tools (center cancellation) | Poor to fair | Free | Seconds | Easy, but only works on some songs |
| Audio editor (Audacity, a DAW) | Fair to very good | Free to paid | 10 to 30 minutes | Install, plugins, settings |
1. With an AI vocal remover: the best result
An AI vocal remover works in your browser, on a computer or a phone, with any song. With the Tool Corner vocal remover:
- Open the tool and drop your song: MP3, WAV, M4A, OGG or FLAC, up to 10 minutes.
- The AI separates a free 30 second preview, taken from the loudest part of the song, usually the chorus, where the voice is.
- Press Play and switch between Instrumental, Vocals only and Both.
- If it sounds right, get the full tracks for €1.99: the instrumental, the vocals and your own mix, as WAV files. No account, no subscription.
A handy extra for singers: the Vocals and Music sliders. Keep the voice at 20 or 30% and you get a guide vocal to sing along with, instead of a bare instrumental.
How the AI tells the voice from the music
Our vocal remover uses Demucs, an open source model from Meta AI research (MIT license). Its latest version, Hybrid Transformer Demucs, was trained on the MUSDB HQ dataset plus 800 extra songs, where every instrument was available as its own track. After hearing thousands of songs and their separate parts, it learned what a voice sounds like, even when a guitar plays the same notes at the same time.
Researchers score separation in decibels of SDR (signal to distortion ratio): the higher, the cleaner. According to its authors (Rouard, Massa and Défossez, "Hybrid Transformers for Music Source Separation", ICASSP 2023), Hybrid Transformer Demucs reaches 9.0 dB on the MUSDB HQ test set, among the best results of open models.
2. With free tools: the center channel trick
Most free "karaoke" buttons, old audio software and many free websites use the same idea. In a studio mix, the lead vocal is usually placed in the center: exactly the same sound in the left and the right speaker. Subtract one channel from the other, and everything that is identical on both sides disappears.
Before: a stereo mix
After: left minus right
That is also why it so often disappoints:
- The bass and the drums go too: they are mixed in the center as well, as our test showed (0% of the kick and bass left).
- Reverb stays: the echo of the voice is spread across both sides, so a ghost voice remains.
- The result is mono, and it needs a real stereo file: a mono recording, or a song where the voice is not in the center, gives nothing.
Free websites that use AI exist too, but they usually limit the file size, the number of songs or the quality of the download. It is worth trying them for one song. For the best quality, the AI way above lets you hear a preview for free before you decide.
3. In an audio editor: the most control
Audacity (free, Windows, macOS, Linux) offers two options:
- Vocal Reduction and Isolation (Effect, Special): the center channel trick, with settings to protect the low frequencies. Since Audacity 3.5 it is no longer included and must be downloaded as a plugin, and the Audacity team itself recommends the AI plugin instead.
- OpenVINO AI effects, free plugins by Intel: the Music Separation effect runs Demucs v4 on your own computer, in 2 tracks (vocals and instrumental) or 4 (drums, bass, other, vocals). Find it under Effect, OpenVINO AI Effects, OpenVINO Music Separation. It works offline, but Audacity documents it for Windows and Linux only.
Music software (DAWs) such as Logic Pro or FL Studio now include a stem splitter as well. Worth it if you already use one and want to edit the stems further. For a karaoke track, the first way is faster.
Finish the instrumental: trim and fade
Once you have the instrumental, you often only need part of it: no long intro for a karaoke video, a 30 second backing track, or a ringtone. Open it in the audio cutter, drag the start and the end, add a fade out, and download it. It runs in your browser: the file is not uploaded.
6 tips for a cleaner instrumental
- Start from the best file you have: a WAV, FLAC or high quality MP3. A low quality copy of a video adds artifacts the AI cannot remove.
- Prefer the studio version: live recordings have the crowd and the room in every microphone, so the voice leaks everywhere.
- Listen to the chorus: if the loudest part is clean, the rest will be. That is why our preview takes it.
- Backing vocals count as vocals: choirs and harmonies are removed with the lead voice. Keep some voice with the Vocals slider if you want them.
- Heavy reverb can leave a trace: a faint echo of the voice is normal on very wet mixes, and rarely heard once you sing over it.
- Keep your original, and name the files clearly (song, instrumental, vocals).
Can I use an instrumental I made?
For yourself, yes: karaoke at home, singing practice, learning a bass line. Removing the vocals does not make the song yours, though. Publishing a cover, a remix or a video with the instrumental of a song you do not own usually needs the permission of the rights holders, or a platform that has a license for it. When in doubt, check the rules of the platform where you publish.
Going further
- Keep only the voice: the vocal remover also gives the acapella, for a remix or a mashup.
- Cut, trim or remove a part of any song with the audio cutter.
- Making a video of your cover? Turn a short clip into a loop with Video to GIF.