In an age where technology is ceaselessly permeating every aspect of our lives, artificial intelligence (AI) has evolved from being a mere science fiction concept into a transformative force across myriad industries. Notably, its intersection with music—both an art form and a universal language—has begun to fundamentally reshape the landscape of music production. From voice generation to song creation, AI has stepped beyond the realms of our imagination and onto the center stage, enabling unprecedented possibilities in the world of music.
Historically, music has always been closely intertwined with technology. The introduction of electricity transformed acoustic instruments into amplified versions of themselves, the advent of synthesizers gave rise to entire new genres, and the proliferation of personal computers empowered countless artists to bring their musical visions to life from the comfort of their homes. Today, AI is the new frontier, the latest technological innovation taking the baton in this long-standing relay of music evolution.
AI’s capabilities are vast and expanding, with its applications in music production rapidly growing. AI’s ability to generate human-like voices, for instance, opens up new avenues for creativity, allowing producers to incorporate a range of vocal styles and tones without needing a physical vocalist. Beyond vocal synthesis, AI has started to dip its toes into the pool of song generation, autonomously creating melodies and harmonies that would have been traditionally composed by humans.
However, AI’s influence extends beyond merely being a production tool. It has started to make inroads into ideation, aiding in the genesis of novel musical concepts and even contributing to songwriting. AI is also finding its place in music education, with its potential to democratize access to music learning and foster a more inclusive environment for musical growth.
Despite its exciting advancements, the involvement of AI in music production is not without its controversies, particularly when it comes to matters of ownership. Questions around copyright, intellectual property rights, and authorship have begun to surface, calling for careful scrutiny and consideration.
Lets delve into each of these aspects of AI’s role in music production, investigating its current applications, future potential, and the ethical and legal challenges it poses. We will examine case studies, hear from industry experts, and contemplate the future of music in an AI-integrated world. The symphony of the future may just be on the horizon, waiting for the beat to drop.
The Unseen Vocalist: AI In Voice Generation
The human voice, with its unique timbre, tone, and expressiveness, has always been a focal point of music. But now, artificial intelligence is venturing into the realm of voice generation, replicating the intricacies of human vocals with an uncanny precision. This capability opens up a plethora of opportunities in music production and sets the stage for a host of fascinating possibilities.
Imagine a scenario where you can create a new track featuring the vocals of any artist – past or present. No studio, no microphones, no scheduling conflicts, just a computer algorithm accurately emulating the distinctive voice of your desired singer. This is no longer a fantasy; AI-based voice generation has started making it a reality.
But how does it work? Voice generation is essentially a two-step process: analysis and synthesis. First, an AI model is trained on a dataset of audio clips, learning to recognize and understand the unique vocal characteristics of different singers. This phase, often known as ‘analysis’, involves the AI understanding the tonality, timbre, pitch, and other specific voice properties.
Once the AI model has been adequately trained, it can then generate a voice that mimics the learnt vocal characteristics. This ‘synthesis’ phase involves the AI using the learnt properties to create an entirely new vocal track. The sophistication of this technology has reached a point where AI-generated vocals can be virtually indistinguishable from the voices of real artists.
A recent instance of this was when the music streaming service Deezer revealed it was developing tools to detect and remove AI-generated content. This move was a response to an incident where a song featuring the AI-cloned voices of popular artists Drake and The Weeknd went viral. It’s not hard to see the ethical and legal complications that arise from such cases, which we will delve into in a later part of this exploration.
However, these concerns should not overshadow the potential advantages of AI in voice generation. It can help in creating a diverse range of vocal styles, experimenting with novel vocal textures, and even bringing back the voices of artists no longer with us. In the hands of ethical, creative individuals, AI-generated vocals could become a powerful tool in the arsenal of modern music production.
As the famed French DJ David Guetta noted in this BBC article, AI might not replace artists or musicians but could serve as a valuable tool. His experimentation with AI to add an Eminem-style vocal to a song highlights how this technology can push the boundaries of creativity, giving producers a new medium to express their musical ideas.
Voice generation, as a component of AI’s role in music, is just the tip of the iceberg. As we delve deeper, we will discover that AI’s influence extends far beyond emulating vocals; it has begun to influence the entire song creation process.
Harmonies From The Machine: AI In Music Creation
The advent of AI isn’t just revolutionising how vocals are produced; it is reshaping the entire song creation process. Let’s take a trip into the world of AI-driven song generation, where melodies and harmonies are birthed from algorithms and data.
In the past, the notion of a machine generating a full-fledged song, complete with melodies, rhythms, harmonies, and lyrics, would have seemed far-fetched. But today, advanced AI algorithms are challenging this perception. They are now capable of creating songs that, while they may not rival a Grammy-winning piece yet, are certainly impressive and showcase AI’s potential in music composition.
OpenAI’s MuseNet is one such AI system that can generate 4-minute musical compositions with 10 different instruments, and can combine styles from country to Mozart to the Beatles. It is a deep learning model trained on a dataset of MIDI files from a myriad of genres and artists, showcasing the power of AI in generating diverse and creative musical pieces.
But how does the process work? Similar to voice generation, song generation with AI is also a two-step process: learning and creating. First, an AI model is fed with a large dataset of songs. It ‘listens’ to these tracks, learning the nuances of musical composition, such as chord progressions, melody lines, rhythmic patterns, and even the artistry of songwriting.
The second step is the generation of music. Using the knowledge it has gleaned from the training data, the AI algorithm can create a new song. It decides on a chord progression, creates a melody, adds in rhythm, and even writes lyrics if required. The result is an original composition, free from the constraints of human biases and preconceptions.
Take, for instance, David Guetta’s AI experiment. While he used AI to mimic Eminem’s vocals, the technology can also create an entirely new track in Eminem’s style or even merge his style with that of another artist. Imagine a song with Eminem’s lyrical prowess combined with Beethoven’s melodic grandeur – AI makes such unique combinations possible.
These capabilities extend beyond mimicking existing styles. AI can push the boundaries of music, creating styles and sounds that humans might not think of. In the future, we may see entirely new genres born out of AI’s capacity to blend, innovate, and experiment.
AI As Your Musical Muse: Idea Creation, Training, and Education
In the realm of music production, the ideation process can often be the most challenging. A single spark of inspiration can ignite a musical masterpiece. But what if the source of that inspiration could be an artificial intelligence system?
With its capacity to analyze vast amounts of data and recognize patterns, AI has proven to be a valuable tool for sparking creative ideas in music. Artists can use AI to generate chord progressions, melodies, or even entire songs to help jumpstart the creative process. This doesn’t mean that AI is replacing the artist; instead, it can be seen as a collaborative partner, providing a different perspective that can lead to unique musical creations.
Start-ups like Amper Music have already built AI platforms that assist users in creating music. These platforms can generate new melodies and sounds based on parameters like genre, mood, or tempo set by the user. Not only can this inspire musicians during the songwriting process, but it also empowers those without traditional musical training to create and experiment with music.
AI is also making its mark in music education. AI-powered tools can now provide feedback on an aspiring musician’s performance, suggest areas for improvement, or even offer online music lessons. For instance, apps like Yousician use audio signal processing algorithms to provide real-time feedback on a user’s performance. This offers an interactive and accessible way to learn music, particularly beneficial for those unable to access traditional music education.
AI’s role in music education extends to training future sound engineers and music producers. Machine learning models can now analyse and learn from professionally mixed tracks, then provide recommendations on how to mix raw tracks, helping sound engineers in training understand how different elements of a track work together to create the final sound.
AI’s Role in Enhancing Audio Quality
As AI continues to revolutionise music production, it is also paving the way for significant advancements in audio enhancement. In particular, one aspect where AI shines is in the extraction and isolation of specific audio elements from a mixed track, also known as “stem extraction” or “source separation”. A shining example of this application is lalal.ai, an innovative AI-powered platform revolutionising audio processing.
lalal.ai uses AI to meticulously separate and remove different components of a track, such as vocals, drums, bass, piano, electric guitar, and synthesizer tracks. This is an extremely complex task, especially when the components are tightly interwoven in a mixed track. Traditional methods often result in quality loss or residual sounds. However, lalal.ai leverages the capabilities of AI to perform this task with an unprecedented level of precision and without compromising audio quality.
At the core of lalal.ai is a sophisticated deep learning algorithm trained on 20TB of data from various music genres. This enables it to understand the characteristics of different musical elements and distinguish between them, even when they’re blended together. Whether it’s isolating a vocal track for a karaoke night, extracting a specific instrument for sampling, or separating elements for a remix, lalal.ai provides clean and high-quality results.
This AI-powered technology is not just useful for musicians and producers. It can also be beneficial for educators, DJs, remix artists, and karaoke enthusiasts. For instance, music educators can use this technology to isolate certain parts of a song to better illustrate a lesson. Meanwhile, DJs and remix artists can extract and manipulate individual tracks to create new music, and karaoke enthusiasts can enjoy their favourite songs without the original vocals.
With lalal.ai and similar platforms, AI is proving to be a game-changer in audio processing, giving users the tools to interact with music in ways that were previously inconceivable. It’s yet another testament to the transformative potential of AI in the world of music.
The Legal Landscape: AI and Copyright
I am not a lawyer, but the use of AI in creating derivative music raises numerous questions, many of which courts are still grappling with. Intellectual property expert, Louis Tompros, outlines two categories of impact AI has on copyright: the rights AI-generated material itself has, and what rights someone might have that they can assert against AI-generated material.
The question of who owns the copyright to material generated wholly or partially by artificial intelligence is crucial. In the United States, at least, recent guidance from the Copyright Office maintains that works created by AI without human intervention or involvement cannot be copyrighted. The reasoning hinges on the interpretation of the word “author” in the copyright statutes, which is generally understood to mean a human. However, this stance has yet to be fully tested in the courts and is expected to face challenges. There is an allowance for human involvement where the individual creatively selects or arranges AI material, in which case the work may be protected by copyright.
Human copyright owners may have rights when AI creates something. Two major categories of questions arise here: an input question and an output question. Does the training required to create complex AI models infringe on copyright if done without consent? And, if the copyright law gives the owner the exclusive right to create a derivative work based on their own prior work, is something created using AI based on that other work then a derivative work?
In the wake of the removal of the track ‘heart on my sleeve’ by artist “Ghostwriter” from streaming platforms in April, the music industry has found itself confronting the implications of AI’s burgeoning influence. This particular track, which simulated the voices of top artists, Drake and The Weeknd, has rekindled debate on AI’s potential impact on music creators and rights holders.
Tompros suggests that the strongest legal argument for the artists might not be copyright infringement, but violation of the right of publicity. This is based on legal precedent where a musical impersonation of a famous musician violates that musician’s right of publicity, as happened in a case involving Bette Midler and Ford Motor Company in 1988.
The person who used AI to generate the song in question could counter that it’s not copyright infringement. They could argue that inputting the artists’ songs into the AI is a fair use and therefore not a violation. They might also contend that the output is not a copy or derivative of the artists’ works because it is an entirely new and distinct song.
Tompros likens the disruption caused by AI-generated music to the advent of music sampling, though the legal issues are somewhat different and more complex. However, he underscores the need to strike a balance between protecting artists’ rights and fostering innovation. Both economic and creative perspectives must be considered to ensure that incentives for new music creation remain strong.
While statutory amendment to the Copyright Act might be desirable to address these issues, practical considerations make this unlikely. More probably, courts will refine their interpretation of the law as they continue to face AI-related challenges.
Whether AI-generated music is eligible for copyright protection is a question that’s significant to both AI developers and rightsholders. For the developers, they seek recognition of their creative input through copyright protection, while rightsholders fear that copyright protection for AI-generated music might undercut their catalogues and undermine human creativity. Given the costs and complexities of licensing and royalty flows, it might even turn out that AI-generated music, if it doesn’t attract copyright protection, could pose a more significant threat to human-made music.
Current copyright law must adapt to keep pace with these rapid developments, a challenge that involves balancing the interests of AI researchers, musicians, and rightsholders.