Workshop Making music with AI

Edited by c’t and Rob Coenraads

	Member benefits This article is from the special Everything about Artificial Intelligence. As an HCC member you can buy these at a nice discount and without shipping costs. Click here for more information and member benefits.

Of all the ways AI is impacting music production, the way it can model vocals has the biggest impact – and that trend is still in its infancy. When it comes to AI’s ability to imitate famous artists, it’s really about one thing: money. And that of course causes problems in almost all cases. We are now in a kind of calm before the storm. There have been notable cases where AI models of famous singers have made headlines. For example, the band Breezer used an AI-generated imitation of Liam Gallagher (including ex-singer Oasis), which the singer was complimentary about, but the anonymous AI-generated Drake song was blocked by Drake’s record company. However, at the moment, AI modeling services make money by selling the use of famous singers’ models, and few of those websites talk about whether the artists in question get a share of the profits.

Royalties

There are only a few websites (such as Kits) that list a royalty split with the original artists on whom the AI ​​algorithms are modeled. Moreover, the artists of that service also have the final say on commercial releases. Another site, Musicfy, also makes clear that it is aware of the ethical concerns raised by using AI models of real, well-known voices. “The potential applications of AI voice cloners range from entertainment to accessibility,” the site said. ‘However, its use requires careful consideration due to the ethical implications. It is essential to to use this technology responsibly, with the understanding that with great power comes great responsibility.’ So it feels a bit like everyone is twiddling their thumbs until there is some legal clarity or perhaps even a high-profile lawsuit to move forward. Suppose you are such a well-known artist, would you be happy if your voice was used in a song, speech or other form of expression without compensation?

Thinking artists

While we wait for that possible outcome, the artists are not sitting still. American singer and composer Holly Herndon uses and promotes AI, even offering a model of her own voice, Holly+. Canadian singer Grimes has also offered help to artists and fans who want to use a model of her voice, as long as there is a division of royalties. And you can’t help but feel that this is a solution that everyone is happy with, with the artists on which the AI ​​model is based getting a share of the revenue. Whether there is a market for albums by AI artists is another matter. Should we have one? could sell a self-made Beatles track – perhaps under the name YesterdAI by the BeAItles? Probably not. But if done well and conscientiously, there could be a market – or at least generate a lot of curiosity – for ‘new’ music from bands that are long gone or no longer have the desire to come up with anything new. AI websites that create singing voices can be divided into two: sites that offer simple text-to-speech with a standard model or models from well-known artists, and sites that offer singing. For the latter, there are plenty of new sites popping up all the time and we’ve also discovered that many of them also offer other services such as basic text to speech, music creation and even AI instrumentation.

Music making workshops

Here you will find two ways to make music. First of all, we explain step by step how you can simulate and try out famous voices with Lalals.com (free). We will then use Musicfy.lol to show you how you can create complete pieces of music, among other things.

Famous voices with lalals.com

From the growing number of websites that sing in offers different styles, Lalals.com seems to be there do everything possible to create simulations of famous ones artists to create with AI, with photos on the opening page. You can try that out for free.	You must first create an account. Then you must choose a subscription model. We chose the free option that allows you you can make three singing parts at a time 20 seconds can be processed

With the free service you can create a singing part make in the style of Ariana Grande, Travis Scott or Drake. You have to make your own for that upload vocal recording. That must be one be an a cappella recording, without any editing or effects, or instruments.	After you upload your song file, go the machine learning at work. It fits the singing style of the selected artist on your recording. A five second recording It was processed quickly, it was done within a minute completed.

Afterwards you can listen to the voice. We were quite impressed with the result. The voice was lifelike and free of artifacts. You can share the results on your social networking, but that’s when you have to upgrade to download it	Lalals.com is one of the simplest websites that you can use and has very many singing models to choose from, including many famous artists and a number that are clearly intended for comic applications

Make music with Musicfy

Musicfy.lol doesn’t focus on that much well-known artists, but more on making parodies and complete pieces of music with using AI. Yet you can too artists download, though it is not clear whether they be based on real ones famous artists. You can also use the service edit your audio for free.

You must first create an account or use a Google account. You can choose from four options: Artists (for singers in different styles), Parodies (for famous ones to vote), Instrumentals and Music AI (for making complete tunes). Then you upload one audio file of a voice, as unedited as possible.

You can then choose from a selection of famous people and characters. Like us said, the famous singers are hidden in the section Artist Vocals; the Parodies algorithms are more about creating comedic singing with famous voices, but we wanted that give it a try anyway.	Press the Remix button and your uploaded voice adopts the sound of the chosen celebrity. A 12 second clip was cut into 10 seconds processed and before we knew it, we had a pretty good Trump impressionist that our sang part. There was some distortion at the high notes.

Many sites offer the option of songs to be made with the help of AI. At Musicfy.lol that’s possible too. Choose Music AI from one of the four options on the left and write a music style you want. We opted for ambient techno and country. It took 40 seconds to complete a segment of 15 seconds to make.	The country AI music was good and better then we’ve heard from other sites, but the techno music was confusingly chaotic and of low quality.