From song file to sing-along track
- Find the best copy you have. Dig out the cleanest version of the song: the album release rather than a live bootleg, and a high-bitrate MP3, WAV or FLAC rather than a file that has been converted three times.
- Open the karaoke maker and add the file. Go to the karaoke maker and drop the file in. A music video file you own works too; only its audio gets processed.
- Let the model split the song. RemoveVocal separates the voice from the music on InverseAI's servers. A progress bar shows where it is. A typical song usually finishes within a few minutes, depending on its length and the queue.
- Audition the hard spots. Flip between Original and Instrumental in the player to judge the split, and play the whole song, not just the chorus: check the quiet intro, the long note that closes the chorus and the stripped-down bridge, since leftover voice shows up there first.
- Save the instrumental. Download instrumental.mp3. That file is your karaoke track. The vocals.mp3 file comes along too, which is handy if a singer wants to learn the melody first.
Why make your own
Commercial karaoke versions are re-recordings. Some are great, some swap the guitar tone or nudge the tempo, and plenty of songs never get one at all. A track split from the original keeps the exact arrangement your singer grew up with: the same drum fills, the same key change, the same fade-out. For a rehearsal that matters, because your timing cues come from the record itself.
Pick the right version before you start
The model can only work with what is in the file. A few minutes spent choosing the source pays off more than anything you do afterwards.
- Studio over live. Crowd noise and room echo get tangled with the voice, and some of that tangle ends up in the backing track.
- Original over remaster-of-a-rip. If the file came from a phone recording a speaker, expect a rougher result.
- Highest bitrate wins. When you own the same song as a 128 kbps MP3 and as a WAV, use the WAV.
- Full length. If you only need the second verse onward, split the whole song anyway and trim it later in any audio editor.
What a two-stem split gives you
RemoveVocal produces exactly two files: vocals and instrumental. Every sung part, including backing vocals and harmonies, goes into the vocals file. For a karaoke night that is usually what you want, because the singer carries the whole tune. For a cover band that relies on the record's harmonies as a guide, it means those harmonies leave with the lead, so plan to sing them live or keep the original playing quietly in one ear during practice.
Tip for rehearsals: keep both files. The instrumental is your practice bed, and the vocals file is a quick way to check a lyric or a phrasing you are not sure about.
Judging the result by ear
Do not listen to the whole song from the top. Jump to the sparse moments. A final chorus with the full band will hide almost anything, but a breakdown with just piano and voice is where a faint ghost of the singer can linger. If you hear it, ask whether it matters: once someone is singing into a microphone over a PA, a soft trace of the original is usually buried. If it is distracting, a cleaner source file is the fix that works most often. Our page on why vocals bleed through walks through the usual causes.
Getting the track ready for the night
- Rename the file to something a tired host can find, like song title, artist and the word karaoke.
- Put all the night's tracks in one playlist so nobody fumbles between apps.
- Play each one through the actual speakers once. A bass-heavy instrumental can sound different on a party PA than on headphones.
- Keep the lyrics on a separate screen or printed sheet; the MP3 carries audio only.
Keep it on the right side of the rules
Stick to songs that are yours or that you have permission to use. Singing over a backing track in your living room is one thing; posting a cover video or playing it at a paid gig may need a license from the rights holders. When in doubt, check before you publish.