Author: Yuval Mann
Translation: Katerina Bubnov
The music world is witnessing a radical transformation by artificial intelligence (AI) infiltration. From an AI-rendered duet by Drake and The Weeknd to an uncanny reunion of Lennon and McCartney, AI is dramatically reshaping our understanding of lyrics and images. As technology evolves, we face an industry-wide revolution characterized by virtual artists in studios and songs generated solely by software. It's an era that may push musicians to redefine their craft.
A peculiar occurrence was observed in Paul McCartney's single "New," where John Lennon's voice unexpectedly resonates from the speakers, sending a chill down your spine. Impossible, one would think, given that Lennon was assassinated in 1980, and the song was released in 2013. The truth is, it's an AI-enabled version constructed by a fan who integrated Lennon's voice samples into McCartney's original track, crafting a Beatles reunion of sorts in the world of AI.
This phenomenon of AI-based songs is becoming increasingly common. In April, a TikTok user @ghostwriter977 released a track, 'Heart on My Sleeve,' purportedly a collaboration between Drake and The Weeknd. Unlike the McCartney and Lennon duet, this was an entirely new song that the duo had never recorded. AI integrated the two singers' voices into the original track.
Despite its AI origins, the song rapidly garnered immense popularity, becoming a viral sensation with millions of streams across various platforms. Drake's record label, Universal Music Group, was less thrilled and demanded its removal from TikTok, Spotify, Apple Music, YouTube, and other streaming services. Yet, in the digital age, the song remains readily accessible despite concerted efforts to erase it.
A simple YouTube search combining your favorite artist or band's name with 'AI' will yield numerous AI-based renditions. Want Kurt Cobain to sing Radiohead's 'Creep'? Or The Beatles performing Queen's 'Bohemian Rhapsody'? Or perhaps a Beyoncé cover of Rihanna's 'Umbrella'? All this and more is now possible. Even Joe Biden performing Neil Young's 'Heart of Gold,' although that's a performance we may not wish to repeat.
Blue Dobrecky, the COO of the AI company Aiode, explains that this is facilitated by a not-so-new technology known as 'Style Transfer.' This tech can reproduce voices and replace the original singers or speakers in existing audio files. While the technology isn't novel, what's changing is its accessibility. Once a tool reserved for tech companies and seasoned programmers, anyone with basic programming skills can use it today.
"Awareness of these tools among users is also increasing," says Dobrecky, a point underscored by the recent wave of AI creations sweeping YouTube.
Dobrecky explains that the technology has now advanced to a stage where a three-second voice recording can generate a speech identical to the person's voice. However, he admits that the magic dissipates when poetry is involved.
"At this point in technology's development, anyone would be able to recognize that it's not a real person," he adds.
Nightmare or Opportunity?
The advent of AI in the music industry poses a difficulty for artists and record labels. The technology poses a severe challenge to copyright protection, as artists like John Lennon, Drake, The Weeknd, and Kurt Cobain never permitted their voices to be replicated and included in creations they were never a part of. But there are concerns beyond monetary exploitation. These include misleading fans unable to discern between authentic and AI-generated works and the potential for AI to insert messages that artists may disapprove of into their songs.
This week, the musician Sting voiced his concerns: "The basic elements of music belong to us, the humans. We'll be fighting in the coming years to protect our human capital against AI." He conceded that AI could be viable for electronic dance music but questioned its efficacy in emulating emotive tracks.
On the flip side, the industry has a significant financial opportunity. Deceased artists like Amy Winehouse, Michael Jackson, or Freddie Mercury could, theoretically, 'return' to release another hit or even an entire album. All this, of course, would need to respect the wishes of their estates and uphold their artistic vision. If there are new songs, why not launch them during a virtual concert tour in the Metaverse or a holographic show, much like ABBA recently did? Neil Tennant of the Pet Shop Boys even suggested AI could help him complete unfinished songs.
One recent application of the technology to resurrect the deceased was by the 11כאן Broadcasting Corporation. They used Aiode's technology to recreate the voices of Ofra Haza and Zohar Argov for a duet called "כאן לעולם" in celebration of the 75th Independence Day. They achieved this by employing vocal imitators and overlaying their performances with synthesized voices of the late singers, all done with the approval of the deceased singers' families.
The singer Grimes, also known as Elon Musk's ex-wife, sees the potential in AI. Following the controversy over Drake and The Weeknd's AI-generated duet, she announced a 50-50 profit-sharing agreement with creators who used her voice to create new music. She even launched a website that allows for easy replication of her voice. This could substantially increase her music-related income without straining a vocal cord. However, this solution may not be suited to most artists.
Dr. Yoav Mor, CEO of Sphereo, cautions: "It will be incredibly challenging for traditional musicians to stand out in a world where a machine can generate a thousand new songs every second. Musicians will need to reinvent themselves and be unique. Only the best will survive."
Daniel Ek, The CEO of Spotify, aptly captured the perplexity and mixed sentiments prevailing in the music industry. Eck coined artificial intelligence (AI) as "cool and scary." He revealed that the streaming service endeavors to strike a delicate balance between fostering innovation on the platform, potentially resulting in revenue-generating new music, and safeguarding artists' copyrights. Ek expressed optimism, emphasizing that AI could significantly enhance creativity in the industry.
Until now, AI's role in the music industry has been limited to altering singers' voices. But AI is expected to play a more significant role in the creative process in the coming years. Companies like Google and OpenAI have introduced models to generate new music pieces based on textual commands, like the ChatGPT and DALL-E2 models that create text and images.
While the technology is still nascent, the results aren't always coherent, and the vocals often sound like gibberish with subpar sound quality. But it's undoubtedly that the technology will improve, and in a few years, AI-written works could serve as background music in video games or films and TV series. We may even reach a point where we can ask AI to compose a new Beatles song or a Bach piece.
Slash, the Virtual Version
Aiode stands out as a pioneering company at the forefront of pushing the boundaries of artificial intelligence in the music industry. As mentioned earlier, they have achieved remarkable success in creating a duet that showcases the combined talents of Ofra Haza and Zohar Argov. Another noteworthy contender in this domain is Mahod Hasharon, a small startup dedicated to developing a platform for virtual musicians driven by the power of artificial intelligence. These virtual musicians are meticulously crafted to emulate real-life artists and possess the extraordinary capability to play and seamlessly adapt to any musical context in real time.
Idan Dobrecki, CEO of Aiode, expresses his company's position:
"We aspire to be the ones who, at least in the music world, provide tools to creators rather than replace them. We involve the musicians in the creative process, get their approval, and ask them which data they'd like our model to train on. Naturally, we share the profits with them at the end of the process."
Imagine an Israeli music producer wishing to record a guitar solo by Slash. With Aiode's platform, they could collaborate with a virtual representation of the renowned guitarist that should sound exactly like him, provided Slash agrees. This highlights Aiode's difference from other AI models trained on millions of works online without permission from rights holders or profit-sharing agreements.
"We have made it our mission to provide tools to creators in the music industry without seeking to replace them," clarifies Idan Dobrecki.
He emphasizes that they prioritize artists' preferences, ensuring that anything they dislike is not incorporated into the model from the outset. Furthermore, after the process, they make it a point to share the profits with the artists - an aspect that, unfortunately, is lacking in models such as DALL-E2 and Midgerani, to our profound disappointment.
Sphereo, another notable startup exploring the captivating crossroads of music and artificial intelligence, has made waves with its latest innovation, Mixit. Dr. Yoav Mor, the CEO, and Founder of Sphereo, describes the app as a complete recording studio within the confines of a mobile device. This cutting-edge karaoke application empowers users to effortlessly transform the musical style of well-known songs at the mere push of a button. For instance, one can shift from heavy rock to a soothing piano ballad or seamlessly transition from pop to an unplugged acoustic rendition. Once the desired transformation is complete, users can record and film themselves singing, enhancing their performances with various filters and effects. The final masterpiece can be effortlessly shared across platforms like TikTok and more.
By integrating with ChatGPT, Mixit users can request the popular chatbot to write new lyrics to familiar songs.
"We're quickly approaching a world where it will be hard to tell the difference between a work created by a human and a work created by a human collaborating with AI," says Mor. "Even though it seems like we're in the early stages, AI's ability to generate music from nothing will arrive faster than we think."
The Implications for the Music Industry: What Are Your Thoughts?
Suppose AI can generate works that rival or match human creations regarding lyrics, music, and performance. In that case, we are headed toward a world where distinguishing between something entirely synthesized and crafted by humans will become increasingly challenging for listeners. In a world where machines can produce a thousand new songs per second, flesh and blood musicians will face significant difficulties standing out. To thrive, musicians will need to reinvent themselves, showcasing exceptional uniqueness and originality. Survival will be reserved for genuinely unique talents.