Guide
How to Remove Vocals From a Song (Make a Karaoke Track)
Turn any song into a karaoke track or extract an acapella using AI stem separation — free, in your browser, no upload. Covers how it works, what to expect, and the limits.
TL;DR
- Removing vocals from a finished song means splitting it back into its original ingredients — vocals, instrumental (or drums/bass/other) — using an AI model. It's not a simple filter.
- Free way: Vocal Remover runs an AI model called Demucs entirely in your browser via ONNX Runtime Web — the song file is never uploaded.
- Choose 2-stem (Vocals + Instrumental) for the fastest karaoke/acapella case, or 4-stem (Vocals, Drums, Bass, Other) if you need to remix individual elements.
- Quality is very good on most modern mixes but not perfect — dense, heavily-mastered, or heavily-compressed tracks can leave faint bleed between stems.
- This is a heavy, memory-intensive process best done on a desktop browser — the tool warns on mobile and blocks very long tracks to avoid crashing the tab.
"Remove the vocals" sounds like it should be a simple audio filter — turn a dial and the singing disappears. It isn't, and understanding why explains why AI-based tools got so much better at this than the old tricks. A finished song is a single mixed-down file where every instrument and voice occupies overlapping frequencies at the same time; there's no separate "vocal track" left inside an MP3 or a WAV to simply mute. Making a karaoke track or extracting an acapella means reconstructing what those separate ingredients probably sounded like — a job modern AI models have gotten remarkably good at.
Why This Is Genuinely Hard (and How AI Solves It)
The classic "vocal remover" trick from decades of karaoke machines and DJ software works by phase cancellation: subtract the left channel from the right channel, and anything mixed identically and centered in both — often the lead vocal — cancels out. It is a neat trick, but a blunt one. It also removes bass, kick drum, and anything else panned dead-center, has no effect on reverb or backing vocals that are not perfectly centered, and does close to nothing on modern mixes that use any stereo widening on the vocal. Vocal Remover uses a different approach entirely: Demucs, a neural network trained on thousands of songs to recognize what vocals, drums, bass, and other instruments actually sound like, running through ONNX Runtime Web directly in your browser. Instead of a mathematical trick on the stereo channels, the model listens to the mix and reconstructs each instrument as its own track — the same category of technology used by professional remix and karaoke software, running locally on your device instead of a studio server.
How to Remove Vocals From a Song, Step by Step
- 1
Open the Vocal Remover tool
Go to Vocal Remover on WeConvertIt. Everything runs in your browser — no sign-up, and the song is never uploaded to a server.
- 2
Drop in your song
Drag your MP3, WAV, or other audio file onto the tool, or click to browse. Any typical song-length track works.
- 3
Choose 2-stem or 4-stem
Pick 2-stem (Vocals + Instrumental) for the standard karaoke or acapella case. Pick 4-stem if you want Drums, Bass, and Other split out separately for remixing.
- 4
Wait for local processing
On first use, the Demucs model (around 170 MB) downloads once and is cached by your browser. A progress bar tracks elapsed and estimated remaining time — typically a few minutes on a desktop computer.
- 5
Preview and download your stems
Listen to each separated track with the built-in player, then download the Instrumental (for karaoke), the Vocals (for an acapella), or all four stems if you chose the 4-stem option.
Getting the Best Result
- Start from the highest-quality source you have — a lossless WAV or FLAC gives the model more detail to work with than a heavily-compressed low-bitrate MP3.
- Use a desktop browser rather than mobile — separation is memory- and compute-intensive, and the tool will warn you or block very long tracks on lower-memory devices to avoid crashing the tab.
- Expect denser, more heavily mastered, or more heavily compressed mixes (a lot of modern pop and EDM) to leave slightly more bleed between stems than a sparser mix with clear separation between instruments.
- If you only need an instrumental, 2-stem mode is faster to download (fewer output files) — processing time is identical either way, so choose based on what you actually need afterward.
What to Do With Your Stems
The Instrumental track works directly as a karaoke backing track — play it back with lyrics on screen, or drop it into any playback app. The Vocals track works as an acapella for remixing, mashups, or practicing a cover. If you separated 4 stems, you can remix individual elements — swap in a new bass line under the original drums, or isolate just the drum stem for a sample. Once you have your stems, Trim & Cut Audio is useful for cutting them down to just the section you need, and Audio Converter can convert the WAV output down to a smaller MP3 for sharing — see the audio compression guide for that step.
Frequently Asked Questions
How do I remove vocals from a song for free?
Use an AI stem separator like Vocal Remover — drop in the song, choose 2-stem, and it splits out a Vocals track and an Instrumental track. It runs locally in your browser with no upload, no account, and no watermark.
Does an old-school "vocal remover" (phase cancellation) actually work?
Not well, and it never really did. The classic trick — subtracting the left and right channels from each other — only removes audio that is perfectly centered and identical in both channels, which happens to include a lot of vocals in older mixes, but also removes bass, kick drum, and anything else mixed dead-center, while leaving reverb and backing vocals untouched. Modern mixes with any stereo processing on the lead vocal defeat it almost entirely. AI-based separation analyzes the actual sound of each instrument instead, and works far more reliably across mix styles.
Can I make a karaoke track from any song?
In practice, yes — the AI model works on the finished stereo mix and does not need the original studio stems. Results are consistently strong on typical pop, rock, and hip-hop mixes; very dense, heavily-compressed, or unusually mixed tracks can leave a faint trace of vocal behind, but a usable karaoke or instrumental track is achievable for the vast majority of songs.
What's the difference between 2-stem and 4-stem separation?
2-stem splits a song into Vocals and Instrumental — the two tracks most people want for karaoke or an acapella. 4-stem goes further, separating Drums, Bass, and Other (everything else — guitars, synths, strings) as individual tracks, useful for remixing or isolating a specific instrument. Processing time is the same either way, since the model analyzes every instrument regardless; 4-stem just means more output files.
Will the vocal removal be perfect?
Very good on most modern mixes, but not flawless — this is a known limit of AI source separation in general, not specific to any one tool. Dense arrangements, heavy mix compression, or vocals that share a lot of frequency space with other instruments can leave faint bleed (a trace of vocal in the instrumental track, or vice versa). It is usually subtle enough not to matter for a karaoke track, but a trained ear can sometimes catch it.
How long does it take to remove vocals from a song?
A few minutes for a typical song on a desktop computer, depending on track length and your hardware — the tool shows a live progress bar with an estimated time remaining. The AI model itself (around 170 MB) downloads once on first use and is cached by your browser after that, so later separations skip the download.
Is it legal to remove vocals from a song and use the instrumental?
Removing vocals for personal use — practicing, singing along, private remixing — is generally not a legal concern. Publishing, distributing, or monetizing the resulting instrumental or acapella of a song you do not own the rights to is a separate copyright question, and the answer depends on the specific song, your jurisdiction, and what you plan to do with it. When in doubt, that is a licensing question, not a technical one.