Lead and backing vocal splitters compared: NiceVois, LALAL.AI and UVR

We put the same song through NiceVois, LALAL.AI and UVR, and published every result below so you can listen for yourself.

Listen

Start with the song, then compare one part at a time. Within each part, the waveforms follow the same moment, so switching tools plays the same bar. Click any waveform to jump. Headphones make the differences much easier to hear.

The song

Lead vocal

Listen for harmonies left underneath the lead, and for gaps where the lead drops out.

Backing vocals

Listen for pieces of the lead left behind, and for harmonies that went missing.

Instrumental

Listen for traces of the voice left in the music, and for music that sounds thin or watery.

What we heard

NiceVois. The lead holds together from start to finish, with no gaps. It sits in the middle the way one voice should, because most of the doubles have been moved out of it. The trade-off is in the backing file. In the quieter moments you can hear a faint trace of the band behind the harmonies, mostly the low end, and here and there a little of the lead.

LALAL.AI. The lead is very close to ours, just a little wider, with a touch more of the doubled parts left in. The backing vocals are the cleanest of the three: almost nothing of the band gets through. They also sound a little softer and less bright than the other two. In the first few seconds, it treats part of the vocal as backing where the other two keep it in the lead.

UVR. Its backing file goes completely silent whenever nobody is singing, which is tidy. The lead is where it struggles on this song. It cuts in and out, and from about 1:30 to 1:55 much of the lead ends up in the backing file instead. It also leaves more of the harmonies in the lead. UVR has many models to choose from, and a different pair might do better on this song.

Instrumentals. LALAL.AI’s is the cleanest: we found practically no voice left in it. UVR’s is close behind, with nothing you can hear. Ours is clean from about ten seconds in, but a very faint trace of the voice is left in the opening seconds.

How we tested

• The same file went into all three on 18 September 2026. Each tool ran once, with its default settings.
• NiceVois: the free splitter on this site, with Also split lead and backing vocals ticked, exactly as any visitor would use it.
• LALAL.AI: the Lead & Back Vocal Splitter, downloaded as MP3.
• UVR (Ultimate Vocal Remover): the same two steps as its Vocal Splitter. BS-Roformer-Viperx-1297 takes the vocals out of the song, then the Karaoke MelBand Roformer by aufr33 and viperx splits them into lead and backing. Both are in UVR’s own model list. We don’t have the UVR app installed, so we ran those model files with audio-separator, a free, open-source tool that runs UVR’s models.
• Nothing was touched afterwards. No volume changes, no trimming, no clean-up. LALAL.AI’s files are exactly as downloaded. The song itself and the NiceVois and UVR results were converted to MP3 at about the same quality as LALAL.AI’s, so the page loads quickly. If one sounds louder, that is how it came out.

The practical differences

As of 18 September 2026, from each tool’s own website and our test.

Which one to use

• LALAL.AI if you want a polished app on every device, and more than lead and backing: it also separates drums, bass, guitars and more. It is a mature product that many people rely on, and on our song it gave the cleanest backing vocals and instrumental.
• UVR if you have a good graphics card and like control. It works offline, costs nothing, and lets you choose from a large library of models and settings. Much of today's separation work, ours included, grew from the community around it.
• NiceVois if you want lead and backing split for free, in a browser, with no sign-up and no graphics card. And if you want to go on and sing the part yourself: NiceVois can train a model of your own voice and put it on the backing track.

What NiceVois runs

Our splitter is built on open models from the separation community, and we think you should know that. The vocals come out of the song with Gabox’s fv7z. Lead and backing are then split with anvuew’s BS-RoFormer karaoke model. After that, one step of our own moves layers that are spread wide in the stereo field, usually doubles, out of the lead and into the backing.

We did not train those models, and we thank the people who did. What we add is running them for you on our own graphics cards, for free, with nothing to install.

Common questions

Is there a free alternative to LALAL.AI for splitting lead and backing vocals?

Yes. NiceVois splits lead and backing vocals for free in your browser, with no sign-up and no limit. UVR is also free if you install it on your own computer. On our test song, the NiceVois lead came out steadier and the LALAL.AI backing vocals came out cleaner.

Which lead and backing vocal splitter is best?

It depends on the song, and one song is not a verdict. On ours, NiceVois gave the steadier lead vocal, LALAL.AI gave the cleaner backing vocals, and the UVR lead dropped out in places. Every file is on this page so you can judge for yourself.

Sources

Every fact about the other tools comes from their own pages, checked on 18 September 2026. The listening results and timings are from our own test.

LALAL.AI

• Pricing: plans and prices
• Lead & Back Vocal Splitter: the feature, its free preview and the monthly price
• Launch post for the Lead & Back Vocal Splitter: the four files you get, and De-echo
• FAQ: the files it accepts
• Output format selection: choosing another format is for paid plans
• Desktop app: an account is needed, and lead and backing splitting runs in their cloud

UVR

• Ultimate Vocal Remover on GitHub: free and open source, supported systems, and graphics card advice
• UVR releases: the current version and its Roformer beta build
• UVR’s model list: BS-Roformer-Viperx-1297 and the Karaoke MelBand Roformer by aufr33 and viperx
• audio-separator: the open-source tool we ran UVR’s models with

The models NiceVois runs

• Gabox’s MelBand Roformers, including fv7z, which takes the vocals out of the song
• anvuew’s BS-RoFormer karaoke model, which splits lead from backing