AIVA Technology – Composing Music using AI
Is an artificial intelligence virtual artist able to compose emotional music?
Founded in February 2016 by three musician-entrepreneurs, AIVA technology is a Luxembourg-based deep-tech startup that leverages AI to create music. The company’s main product AIVA – the Artificial Intelligence Virtual Artist – is able to compose emotional soundtracks for movies, commercials, games, trailers, and TV shows. The mission of the company is “to establish AIVA as one of the greatest composers of all time, and fuel the world with personalized music”. In fact, AIVA is the first-ever AI in the world to officially receive the status of Composer from a rights society, and formally holds the right to own the copyright and receive royalties for the music it composes.
When it first started, the company focused on developing classical symphonic music for films. Today, AIVA can compose soundtracks in a variety of genres, ranging from electronic to jazz, pop, rock, ambient and exotic styles such as Indian and Chinese, enabling artists to collaborate and compose soundtracks of their desired style. The company serves both individual artists and institutional customers, such as advertising and production agencies. Its goal is to make the music production process easier, faster and smoother without compromising the quality of music.
How does the technology work?
The main technique employed by AIVA is deep learning. Specifically, AIVA reads through a large database of music written by the most famous composers in history, and looks for patterns and rules in the scores to understand the style of the music. According to the company, “AIVA practices its understanding of music by predicting what comes next in the track. Once it gets good at these predictions, it creates a set of mathematical rules for that style of music”. By doing so, AIVA can capture concepts of music theory and produce melodies that are appealing to human ears just by learning from existing musical works. This process of having AIVA learn from trials and errors is called reinforcement learning, a core technique in machine learning.
But if the algorithm is only based on extracting patterns and replicating them, wouldn’t the music all sound similar? In order to create unique elements, AIVA uses deep neural networks to generate musical variations. For a given musical theme, AIVA iteratively applies changes to the harmony, melody and rhythm of the piece. While staying faithful to the original theme, it gradually moves away from it with each iteration based on the specified preferences indicated by the customer. In this way, the music is personalized to the customer.
AIVA has been secretive about the specific algorithm it uses to compose music, so certain aspects of the algorithm remain mystical to me. For example, it is unclear whether and to what extent human-defined schemes still play a role in the musical works created by AIVA. Does the neural network completely randomize and restructure the input music in order to produce a new composition, or are human judgement involved at certain stages of the production?
Pricing
AIVA’s pricing structure is a 3-tier licensing model. In the bottom tier, the service is completely free; all copyrights for the generated music belong to AIVA, and the user has no right to monetize the music. Those who subscribe for service at the cost of $35 per month, however, own the copyright of the music and are free to monetize without any restriction.
Opportunities and Challenges
In the past few years, the company has been growing steadily. It has raised more than $1.7 million of capital through VC and angel investors from around the world, and has released several albums, including the classical track “Genesis”, the soundtrack for the video game “Pixelfield”, and a variety of other creations.
The AI music industry is growing. AIVA faces competition from other actors such as Amper Music and GliaCloud. Being able to attract enough subscribers is crucial to AIVA’s growth. In my view, AIVA should position itself as a music incubator, and target advertising and production agencies, not just individual customers. After all, music producers are a specific set of customers to serve, but the entertainment industry is larger, with more ample, varied and sustainable demand for the type of service offered AIVA.
In addition to competitive pressure, AI music faces its unique challenges. When a AIVA-composed song was played at the National Day celebration in Luxembourg in 2017, a newspaper review described it as “an affront to Luxembourg composers, a slap in the face of all creative people in the arts.” Controversies around whether AI music impedes creativity and eliminates jobs for human composers abound. On these matters, AIVA emphasizes that their goal is not to replace musicians, but to encourage more collaborations between human and machines. “AI can fix writer’s block, and address use cases that composers alone cannot solve, but still need humans to be brought to life, with real instruments and a conductor’s irreplaceable artistic supervision,” the company says.
AI music is also a nascent industry where regulations fall behind. At some point in the future, the company might need to deal with regulatory risks around copyright issues. The company should anticipate such risks and expand its service beyond mere music production and copyright ownership. Looking to the future, more diverse, creative, and collaborative services add both attraction and protection to AIVA and its entrepreneurial, social, and artistic mission.
Very interesting post, Yifei! It was interesting to read about the competitive landscape. I wonder if this is a segment that the music streaming platforms are well-positioned to enter given the amount of data they have on consumer preferences–and the various patterns they might be able to recognize about trends in musical styles. I agree with your take that AIVA should focus on the production agencies as the end customer. Listening to the sample from Pixelfield, it sounds like AIVA’s product is perfect for a businesses that would otherwise purchase “stock” background music with simple chord patterns.
Hey Yifei Wu. Thanks for the insightful post. I watched Black Panther: Wakanda Forever a few weeks ago, and fell in love witht the sountrack by Ludwig Göransson. It’s hard for me to believe that AI could one day compose atmospheric and breath-taking soundscapes like that. I understand that AI-generated compositions may be used in video games where compositions need to be dynamic and adaptable to user inputs, but there may be artforms like movies where artistic composition is desired. So I defintely agree that AVIA’s technology is not necessarily meant to replace humans. It’s all about the use case!