AI helped the Beatles release one final song

Hey there, AI fans

A lot of stuff has happened in AI in the last weeks, and while much of the hype seems to have died down, the speed of development and news have not… 

OpenAI/ChatGPT released some crazy things this week. It is huge in so many ways! I will cover that next week to report more details about it, what people are doing, and what you can do with it!

Today, I will talk about how AI is revolutionizing the music industry and, of course, cover the “big deal” moment of the Beatles' final song, “Now and Then.” All made possible thanks to AI. 

Ready?

Let’s go! 

Yesterday, I sent out the third edition of the Cancer Innovations - a weekly newsletter from the intersection of cancer and AI. Sounds interesting? 

Sing up here: https://cancerinnovations.beehiiv.com/subscribe 

Beatles release one final song, thanks to AI

43 years after John Lennon’s death, the Beatles released their final song, “Now and Then”. 

The amazing part here is that all of the “Fab Four” participated in the song, despite two of them being dead… 

Here are the main points of the story:

• John Lennon records “Now and Then” (and three other songs) to a handheld cassette recorder perched on his piano in 1977. 

• Due to noise pollution in the room and the lack of professional recording tools, the quality of Lennon’s voice is too bad to use. 

• The three other songs were edited and released at some point previously (better quality/less room noise). 

• The “final song” was attempted to be edited again in 1995, where Georg Harrison played his guitar part. But the work is abandoned after two days. 

• In 2021, the remaining members, Paul McCartney and Ringo Starr, worked with Peter Jackson for their Beatles documentary “Get Back.” Peter Jackson and his team developed an AI that can separate sounds, instruments, and voices from each other from old mono recordings, like the “Lennon tape” from 1977. 

• With Peter Jackson’s LLM/AI, they were able to separate John Lennon’s voice from the tape in 2023 and decided to finish the song. They use Lennon’s original voice from 1977, George Harrison’s guitar tracks from 1995, and Paul McCartney and Ringo Starr do the rest in 2023. Amazing! 

• On 2 November, “Now and Then” was released, and it goes to no 1 in the hit lists. 

I don’t have much information on the AI Peter Jackson and his team developed. However, no voice synthesis was used to recreate John Lennon’s voice. 

Through training, the AI learned the difference between instruments, sounds, and Lennon’s voice. Finally, the AI managed to separate Lennon’s voice, giving them a crystal clear sound file of Lennon singing, allowing “one final song” as a band. 

You can watch this (12 min) documentary on the song's creation. If you just want to hear the results after the AI has done its thing, skip to 07:02. 

Resurrecting the voices of dead artists is just the tip of the iceberg of what’s possible with AI. 

In the case of using AI in music, Paul McCartney has been one the forward-looking artists, yet stated that it is both scary and exciting for music. 

Drake and Ed Sheeran are artists on the opposite side of what is happening. 

Artists lost a lot of revenue with the death of CDs. I think AI is going to make it even harder… 

Musicians will need to go on tour to make money. I guess the flip side of that is that we need to get out of our living rooms, away from the screens, and go out and interact with other people and be social… Maybe AI is not a bad thing after all…? 🙂

🤖 

AI-generated royalty-free music

Tools like Soundraw let you AI generate royalty-free music with commercial rights for a small sum of money. You can select a host of parameters, like the length of the track, tempo, genre, and more, and the AI will generate tunes for you. 

If you create YouTube videos, you can now create unique music for your videos without having to worry about copyright strikes. 

Or, if you have a podcast and need music, you can AI generate it without worrying about licensing. 

You’re a singer but don’t have a band? Now you do… 

If you pay for the highest tier of Soundraw, you can even distribute to Spotify and Apple Music and make money on “your” music. 

Another option is Soundful, which is considerably cheaper than Soundraw (currently $30 per year). 

Mubert is yet another one. 

Beatoven one more. 

While you can find free royalty music online, why not generate something unique to you? Worry-free music… 

You can use AI-generated music for YouTube, Social Media, Streaming Services, Websites, Corporate Videos, Digital Ads, Live Broadcasts, Video Games, Apps, NFT, Eve ts, and even background music for stores (yes, that is a thing! You cannot stream Spotify in your business without a proper license, although 1000s of shops/businesses do so all around the world). There are probably many more use cases than these. 

🤖 

Music Creation with AI

There are also generative AI tools that can generate music with just a simple text prompt. Apps like Boomy are transforming the way music is made. These apps leverage extensive music databases to learn and replicate various elements of music, enabling even nonmusicians to create songs. 

Users can generate music and upload it to Spotify to make money off their creations despite lacking music skills (hey, Mom, look, I’m a musician now). 

Boomy became so popular that Spotify started removing a lot of Boomy tracks back in April/May. All this AI-generated music raised concerns about artists' rights and royalties. A quick resolution allowed the re-uploading of these tracks, yet the potential for copyright infringement by AI remains a topic of discussion.

Streaming giants like Spotify and Amazon Music are also exploring AI music-generation technology. Today, Spotify pays about 70% in royalties to the creators. With all this AI-generated music, what’s stopping Spotify or Amazon from generating their own AI music and keeping 100% of the revenue? 

Many industries are wondering what AI will do to it. The music industry is no different. Bypassing human artists and reshaping the economics and artistry of music creation is one of those concerns.

Is AI music creation like this a good or a bad development? 

Tell me what you think! 

🤖 

Enhancing Music Production with AI-Powered Mixing and Mastering

Mixing and mastering are areas of the music industry where AI has been deployed for years. AI apps like Landr, Cryo Mix, and iZotope’s Neutron are at the front of this revolution. These innovative tools are reshaping the mixing and mastering process, making it accessible for musicians and producers of all levels. 

AI in music production simplifies the technical aspects of balancing audio levels and noise removal. These technologies also empower emerging artists to achieve professional-quality sound without the need for expensive audio engineers. That’s pretty awesome, I think.

Tools like these encourage kids at home to create music and express their talent without needing music labels. 

How many musicians started their music careers from their bedrooms in their parents' houses? With all the modern music technology available, we will see more and more artists like this emerge. 

I’m from Norway, and two huge artists who came from somewhat similar situations like these are Kygo and Alan Walker. I am sure there are many like them from your country too! 

🤖 

AI-Powered Instrumental and Vocal Reproduction

Mawf and Vocaloid are two apps pushing the boundaries of what's possible in music creation.

Mawf's "tone transfer" algorithms enable musicians to convert the sound of one instrument into another, opening up a realm of creative possibilities. 

This technology was notably showcased in the winning entry of the 2022 international AI Song Contest by Thai musician and engineer Yaboi Hanoi. His track incorporated Thai mythological influences and featured a digital recreation of the traditional Thai woodwind instrument, the pi nai.

Similarly, Vocaloid's voice synthesis software allows users to produce strikingly human-like vocal tracks with interchangeable voices. This technology empowers musicians and producers to reproduce any instrument or voice realistically, vastly expanding the scope of musical expression.

I have spoken about voice cloning tools like ElevenLabs, and Deep Fakes was a topic earlier this year. These technologies showcase the darker sides of what AI can do (but there are positives, too!). One of the “TikTok famous” AI generations is the Drake & Weekend AI fakes (this is all AI-generated, but it sounds exactly like Drake and the Weekend).  

Here is another AI fake of Travis Scott:

🤖 

That’s all I had for you today!

I hope you found this interesting and gave you a look into what is going on with AI in the music industry! 

What do you think about the future of music? Are we all doomed, or is this an amazing development? 

AI’ll be back! 

- Thomas

🤘