Musicians have been remixing music from famous artists for decades. Getting a hold of the instrumental and a cappella vocal tracks used to be much harder though. Advances in artificial intelligence are changing the game, giving everyday people access to the audio stems of published music. That's one of the reasons we've seen a surge in AI song covers in 2023.
You probably heard about Heart On My Sleeve, the AI drake cover song that went viral in April 2023. Published anonymously by Ghostwriter 977, the track reached 600k Spotify streams, 275k YouTube views and 15 million TikTok views before being taken down. Several other AI song covers have since made their debut, though few have reached this level of popularity. So how are people doing it?
VocalRemover's Splitter AI service has become the number one web app for isolating vocals and instrumentals online. You can extract the bass, drums, and the rest of the instruments from any vocal tracks. It's fast, free to use, and creates high quality audio that site visitors can download without ever signing up. Other paid services like Lalal.ai and Moises charge a substantial amount by comparison.
In this article, I'll outline some workflows that people have been using to create AI song covers from isolated vocals and instrumental tracks. There are black-hat techniques that undermine artists and put you at risk legally, but there are also constructive ways to use these tools for learning the art of music production. It's on each individual to decide what path to take.
Table of Contents
Legal disclaimer: Don't steal from artists
The legal implications of vocal removal are self-evident. If you steal instrumentals and vocal tracks from published artists for the purpose of publishing new music commercially, you're breaking the law. Sample-based production has been a part of popular music for a long time, but ripping off entire songs is a different story.
We don't advocate for the misuse of any audio software. The livelihood of artists have already been damaged by streaming platforms and predatory record labels. The pandemic took a swipe at touring revenue, which was one of the last places where professional musicians could earn a living. So be nice and support the artists that you love.
If you do plan to use these techniques, keep your ethics in tact. We'll share ideas for how you can experiment and learn from published music without stealing.
Splitter AI: Isolating vocals & instrumentals for free
Splitter AI is a free service offered by VocalRemover.com. Despite the high cost of processing music in this way, they seem to be staying afloat on donations and perhaps some private investments. Here's an overview of how the service works and what people are doing to access the music.
How do you use Splitter AI to remove vocals?
To get started, navigate to the Splitter AI webpage and click on Browse my files to upload a track from your computer. You won't need to sign up to get started or download the final audio. It only takes a minute or two to upload and process audio for a song of average length (2-4 minutes).
When the track is finished processing, you'll be able to listen to the vocals, bass, drum, and remaining instrumentals as separate tracks. The interface includes a basic mixer with volume levels for each audio stem.
As a bonus, Splitter AI also includes the project tempo and key signature. If the song isn't too complicated (no shifts in tonal center or BPM) then these measurements tend to be accurate.
You can use the BPM and key signature data to help you out later when you take these stems into the DAW and begin experimenting.
Click the save button at the bottom right corner of the screen to access the individual audio stems. You can also download the full mix based on the volume levels you set for each track. There's even a STEM option for those of you using Native Instruments.
Obtaining commercial music to use in Splitter AI
This article is focused on AI cover songs, so we're assuming that you're not trying to run a stem splitter on your own music. Here are the two main techniques you can use for uploading other people's music. It's simple - either you pay the artist or you don't.
Purchasing a song and supporting the artist
The most ethical way to practice this technique is by purchasing the audio file that you plan to experiment with. Platforms like Apple Music make it easy to obtain individual songs on demand. By paying for the track, you are contributing to the artist's livelihood. Then you can freely experiment with these techniques in private. You should not sell your revisions commercially. Even publishing derivative content on a platform like YouTube is questionable.
Ripping audio from websites (not recommended)
Ripping audio and video from the internet is a touchy subject. Websites that offer services like this fall in the same category as torrenting platforms. They're often filled with ads and have a spammy feel to them. You might get a virus using them. Authorities work hard to take them down and so they tend to go offline.
You should always support artists by purchasing their music when you can. If you don't have those resources and want to experiment privately, the best workaround is to use free audio routing software like Rogue Amoeba's Loopback with a local audio capture tool like QuickTime.
By setting Loopback's audio source to your browser (Google Chrome for example), you'll be able to capture audio in QuickTime by selecting it as the audio source.
Once you've got the audio router and audio recording software configured, you can navigate to YouTube (or any other website) and hit record. Then press play on the website and grab the audio. When the audio is finished, you can save that file to your local drive and upload it to a service like Splitter AI. Once the vocals and instrumentals are isolated, you can save them and begin to create AI song covers for personal use.
This is kind of a crazy hack. If you're familiar with the Web3 space, you may recall the controversy when people started right clicking and saving NFTs instead of purchasing them. I would put this in a similar camp. You're stealing from the website and its creators when you do this. It's not ethical to circumvent a core principle of trust on the internet and if too many people do it, the incentive to create new cool stuff goes away.
This audio routing technique with Loopback has a second, completely legal and practical application. You can capture audio from your DAW by assigning it as an audio source on your router (Loopback) and then set the DAW's audio output to that router. Here's an example of what that looks like in Logic Pro X:
With this technique you can record screen captures of your work in a DAW, in simple video capture tools like QuickTime as well as free live streaming software like StreamLabs. This is a great way to create tutorials and show people your music making workflows, whether it's about creating AI cover songs or sharing any other music production tips.
Creating AI song covers for non-commercial use
By this point in the article, it's abundantly clear that we don't endorse the misuse of commercial music. The following techniques for creating AI song covers are for personal, self-educational purposes only. If you choose to publish any music using these techniques, you do so at your own discretion.
AI remix: Creating new songs from Splitter AI instrumentals
Now that you have the audio stems for a song, you can start studying how all the pieces fit together. What would it sound like to use a different bass or drum loop?
Thanks to Splitter AI, you already have the project key signature and BPM. With that information, you can go down at least three different paths.
1. Sample marketplaces: Go to a website like Splice and filter by instrument, tempo, and key to find the perfect replacement track. We published an article last week on Splice AI tools that find similar sounds, so if you find a sample that's almost there, but not quite right, this technique will save you a lot of time.
2. AI music generator: There are plenty of AI music apps that create original instrumental tracks based on key and tempo. Import the Splitter AI and AI music stems into your DAW and experiment with the mix.
3. Write the new track yourself: If you're looking to build up your musicianship, try writing your own replacement track. Record audio or MIDI yourself to imagine what you would do if you had been one of the session musicians. This falls loosely into what I call the AI bandmate workflow.
If you want a fun way to experiment with new MIDI ideas for your song, without using AI, try using the AudioCipher VST. The app will encode your words directly into your melodies and chord progressions, like a coded language. You can type in lyrics from the song, choose your key signature, and drag the file to MIDI in your DAW of choice.
New vocals: Using AI voice generators
1. Voice transfer for the isolated vocals: You'll have to check the terms and conditions for an AI vocal service before doing this. For legal reasons, I can't recommend a specific service to do this with. But practically speaking, you can upload the isolated vocal stem and use voice-to-voice technology to hear the celebrity voice performing as a different artist with the same inflections.
2. Voice transfer for your own vocals: Whether you write new lyrics or sing the original words, you can record your voice and use voice-to-voice transfer. A lot of people have used So Vits SVC to impersonate celebrity singing voices.
3. Text to melody: Our AI voice generator article, linked above, includes a few options for DAWs that take your lyrics and give you control over the melodic shape of those words. If you have Melodyne, you can also use text to speech apps like UberDuck for vocal impersonation.
3 Alternatives for generating AI song covers
Here are a few alternative apps you can use if stem splitting and remixing in the DAW sounds too complicated.
Chirp by Suno AI: If all you're looking to do is apply the lyrics to a new parody song, have a look at Chirp. They're a free Discord server that generates AI music with instrumentals and vocals, based on your lyrics.
VoiceMod Text-to-Song: For a second lyric-to-song generator, check out VoiceMod's Musical Meme Machine. It comes with a number of AI voices and a few different styles of music. It has a more polished feel than Chirp, but it's also less imaginative because of being so tightly controlled musically.
Covers.AI: The Covers.AI service, powered by virtual music studio Mayk, is marketed as an easy way to create AI cover songs quickly. Users can sing some lyrics (poorly is okay), upload the file, and hear what it generates.
These options offer less learning opportunities for musicians, but if your goal is to make a quick joke with your friends, they will do the trick. For more ideas on experimenting with AI music, check out the AudioCipher blog.