What is the best vocal extraction software? We compared the top AI technology to find out

Last updated on February 8th, 2023 at 12:29 pm

Sponsored Post

Within the last few years, music creators like myself have seen an influx of AI-based audio processing companies that claim to possess the technology to split audio tracks down to their subsequent parts.

These separated parts, or “stems,” can be highly valuable to DJs and producers who have the desire to remix or sample a song. It eliminates the need for wasting precious minutes either attempting to separate it yourself or finding the track online already separated for you.

I have spent hours of my life doing both… and ultimately both of these archaic options tend to leave you with a less than desirable product.

But Feather, I’m not a producer. I’m not a DJ. Why should I care?

Ultimately stem separation can be extremely useful for the rapidly expanding creator market. Maybe you’re a vlogger or the next TikTok queen.

Talking over audio with vocals can be extremely distracting and detrimental to your content. Removing them could fix that in a second.

Maybe you need an instrumental or karaoke-style backing track for a thrilling and elaborate presentation at work or school, rewriting the words to “Hey There Delilah” to instead inform the audience about the parts of the human brain (I’ve done this).

Or remove the original vocal to save spending hours recording your own instrumental version (I’ve also done this).

Maybe you’re an AC/DC fan who’d love to practice your guitar chops to just the drum track of “Back In Black”, or maybe you just want to make your own bootleg remix of Taylor Swift’s “Love Story” for your sisters Nashville-bound bachelorette party.

This is where stem splitters can help! 

Gone are the days of spending hours digging through subreddit pages, shady unkept websites, and the deepest darkest holes of YouTube just to find a suitable acapella or instrumental for your production.

This is the future baby, where the machines do it all for us. Now, web-based software like LALAL.AI, Splitter.ai, VocalRemover, PhonicMind, and Spleeter by Deezer (to name a few) are on the rise to fix this problem and offer high-quality versatile solutions.

But the question still stands… which one do you use?

Enter LALAL.AI.

What Is LALAL.AI? 

LALAL.AI is one of the latest companies on the scene. Starting in 2020, LALAL.AI has upgraded its audio processing and machine learning algorithms each year, priding itself on touting the latest and greatest tech to make music creation and mixing easier for DJs, musicians, sound producers, engineers, and all creative people.

Fresh off the release of their next-generation neural network “Phoenix” in February of this year, LALAL.AI is confident they can now split your files twice as fast, with significantly fewer artifacts, and an improved focus on vocal extraction.

How Does LALAL.AI Work?

Now, audio processing and splitting data is no easy task. There is a lot of computer science that goes down to make this all work, and frankly even a relatively tech-savvy individual like me struggles to totally gather what’s going on fully.

Nevertheless, I’ve chewed through their blog on the latest updates and backend science to get ya the cliff notes summary.

Using a team of specialists in the fields of artificial intelligence, machine learning, mathematical optimization, and digital signal, LALAL.AI has spent years of research and terabytes of data to “train” a neural network to recognize what types of sounds are specific components of a song. 

Okay, maybe I lied – that was still a bit much. Basically, what you need to know is a bunch of smart guys and gals wrote a bunch of smart code using examples of songs and math (our favorite subject) and put it in a smart computer so when you upload a file, it can split it into different parts! 

Sounds intense and it probably is. But for you to use it? It’s easy peasy (R.I.P. Bob the Brain).

Immediately upon visiting LALAL.AI’s website you are greeted by a landing page with a large spot to drag and drop your audio file. Simply plop it on there, and select from the following dropdown menu what instrumental part you would like separated.

Currently, LALALA.AI supports the separation of drums, bass, electric guitar, acoustic guitar, piano, and synthesizer along with featured separation of vocal and instrumentation utilizing their latest and greatest Phoenix technology. 

After uploading your desired track, LALAL.AI will process the audio and provide you with a sample of what the separation will sound like.

From here you can choose to move forward and process the entire file at which point the total length of the file will be deducted from the time limit on your account and you will be provided a download link upon completion.

Is LALAL.AI Free? 

Currently, LALAL.AI is available as a free service but they do offer two one-time paid pricing tiers for those users who are looking for some power options.

While all packages are restricted by time limits, the free service also is capped at a 50MB file size limit, slow processing queue, and only supports MP3, OGG, and WAV file formats.

Lite and Plus packages cap out at 90 and 300 minutes respectively and have fast processing with a 2GB size limit. 

For most users, I’m sure you can get away with a free account for a majority of projects. The most common file formats to deal with are MP3 and WAV anyway. The typical file size for the WAV of an average song is roughly 40MB so you should be just fine.

Even when testing the service myself, I was in and out in under 5 minutes tops so I didn’t miss the fast processing queue times.

Is LALAL.AI the Best Splitter?

Ah, the legendary question… which is the best tool for the job?

When it comes to things like this I find most people tend to avoid offering a direct opinion on what you should do. They are too afraid to be responsible for the possibility that you may have a bad experience and therefore settle on a safe middle-of-the-road opinion that favors all sides.

I, however, am not that person.  I’ve spent many hours reading reviews or pro/con write-ups about products and experiences only to get the bottom of the article and walk away with no tangible advice on what direction to head next.

That, in my opinion, isn’t helpful. We seek guidance to help sift through this overwhelming world of online information and therefore\ I find it my job to share what I know to make that process easier.

I pride myself on my research. I spend lots of my time looking for the best option for whatever task it is I am doing.

My thought is that if I’m going to do something, I should spend the time learning about it to find the best option – do it once, and do it right.

While I will agree that the best option does depend on what application you’re using it for and a variety of contexts, I’m not afraid to point you in a direction to get started. Because after all, sometimes getting started is the hardest part. 

Lucky for you all, I have needed to separate countless audio files in my ventures over the years. As a captain of the DIY lifestyle, I got tired of trying to find the things I needed already made online and decided to just do it myself.

This was a couple of years ago and AI-based splitters were a newer tool at the time. In the end, my solution was to utilize a program many music creatives may recognize called RX 8 made by audio software company iZotope.

RX 8 features a processing and separation tool very similar to most splitters on the market and it did the job for me as I was mostly separating basic vocals and instrumentals.

At the time, I already owned RX 8 as it came with a plugin bundle so I was in luck, but to buy this software new it will run you about $400 – not cheap.

At one point, I decided to try LALAL.AI per the recommendation of a friend to see if it was a better experience than my current process and I was left feeling that it was mostly the same. Nothing new or exciting. 

HOWEVER, that was a while ago before they had released their latest and greatest Phoenix technology. Now there’s a shiny new toy to test out and compare and you know me, I gotta do my research. 

How Does LALAL.AI Compare?

Eager to test this out, I decided to reference one of my very own songs for comparison. This way, I have access to the original a capella so we can truly compare quality between our options.

Utilizing the master WAV for my song “Let Go feat. Luma” I proceeded to go through the motions of separating the vocal using RX 8, LALAL.AI, and two other separation competitors: Splitter.ai and VocalRemover. The showdown was underway.

Time

The biggest difference here between web-based splitters and on-computer software like RX 8 is upload time.

LALAL.AI, Splitter.ai, and VocalRemover all took about a minute to upload with my internet but obviously, this is subjective based on your internet speed, so I won’t count this much.

RX 8 takes the win here as it requires zero upload time. Just add the song and go. 

Immediately I was impressed with LALAL.AI’s processing time and it took just under two minutes to split the vocal from the instrumental.

Phoenix is looking good.

However, VocalRemover had split the track even faster at around a minute. RX 8 takes some hits on this end as it doesn’t feature the ability to simultaneously split the vocal and instrumental.

It took me over twice as long to process and then reprocess the audio for each part, so VocalRemover takes the cake here. 

As for Splitter.ai, as I type this I am STILL awaiting the confirmation email from them that my audio file is done processing and ready for download – and yes I typed it correctly.

I’m not sure what is going on here, but at this point, they have already lost me as a customer.

My precious time doesn’t care to spend it troubleshooting when there are already handfuls of other options out there to use, especially ones as fast and functional as LALAL.AI and VocalRemover.

One down. 

Quality

The good stuff. After downloading all vocal and instrumental files from each source I added them to my DAW of choice to A/B the files side by side. 

While all versions of the separated vocal contained a fair amount of audible artifacts (digital noise created through the extraction process) this is not necessarily a standard result.

Depending on the instrument you are separating and the instrumentation you are separating it from, you can get entirely different results – so remember that.

In this case, compared to the original track a cappella, all results were average: C+, mid at best.

In my opinion, I would stray away from using any of these in production unless I only needed one small part that did not need to be clearly heard.

However, I know maybe this track just wasn’t the best for this experiment and I don’t hold that against our competitors.

All that being said, when compared side-by-side LALAL.AI’s vocal extraction was clearly the strongest, often exhibiting a more full and round tone and fewer artifacts compared to the others.

VocalRemover’s file often had a thin sandy feel to it and struggled with including background vocals, and RX 8’s result contained the most artifacts by far. 

When it comes to the instrumental, both LALAL.AI and RX 8 had very competitive solutions.

Unfortunately, VocalRemover’s extraction still had a very sandy and processed-sounding file, so at this point, I’ll remove it from the shootout as its quality lacks considerably from the others.

Two Down. 

Depending on the part of the song LALAL.AI and RX 8’s separation could vary. Sometimes vocals would peek through or extra instrumentation would be missing.

It seems that RX 8 is better at retaining background instruments and is more liberal with vocal separation whereas LALAL.AI tends to favor extracting more vocal than a more conservative approach.

Of course, this is where the stalemate lies. Fortunately for RX 8, they have settings you can tweak to dial in the exact type of sound you’re looking for.

While this might take some time and troubleshooting, I believe that it will ultimately end with a better, more quality product.

Personally, as I am a bit of a perfectionist, I’d have to give it to RX 8 in the quality control department. 

So where does that leave us?

Should I Use LALAL.AI?

At the end of the day, I’m guessing most of you reading this don’t have the extra $400 sitting around to invest in some extra software.

Save that money.

You don’t need to spend it just to get some complex software that is only slightly better. In fact, even if you do, I imagine you might not enjoy using it as much as LALAL.AI for your separation needs.

Where LALAL.AI shines is its incredibly fast and efficient process for uploading, processing, and separating audio tracks in one smooth user-friendly experience, ESPECIALLY for vocals and instrumentation, utilizing its new Phoenix technology which will be applied to its other areas of instrumentation later this year.

Despite the more flexible controls of computer-based software like RX 8 that might ultimately yield a slightly higher quality result, I think that is negligible for anyone else looking for an enjoyable experience and quality product.

The improvements LALAL.AI has made in the last year and will continue to make in the future have not only changed my opinion of their service but made it clear that they are the number one option for separating audio files.

Whether you’re an active DJ, producer, vlogger, or just want to make a karaoke track of 50 Cent’s “In Da Club” for your grandmother’s 90th birthday, choose LALAL.AI for the best vocal remover.