MusicLM

Music and artificial intelligence (music and AI) is the development of music software programs which use AI to generate music. As with applications in other fields, AI in music also simulates mental tasks. A prominent feature is the capability of an AI algorithm to learn based on past data, such as in computer accompaniment technology, wherein the AI is capable of listening to a human performer and performing accompaniment. Artificial intelligence also drives interactive composition technology, wherein a computer composes music in response to a live performance. There are other AI applications in music that cover not only music composition, production, and performance but also how music is marketed and consumed. Several music player programs have also been developed to use voice recognition and natural language processing technology for music voice control. Current research includes the application of AI in music composition, performance, theory and digital sound processing. Composers/artists like Jennifer Walshe or Holly Herndon have been exploring aspects of music AI for years in their performances and musical works. Another original approach of humans “imitating AI” can be found in the 43-hour sound installation ''String Quartet(s)'' by Georges Lentz (see interview with ChatGPT-4 on music and AI).

20th century art historian Erwin Panofsky proposed that in all art, there existed three levels of meaning: primary meaning, or the natural subject; secondary meaning, or the conventional subject; and tertiary meaning, the intrinsic content of the subject. AI music explores the foremost of these, creating music without the "intention" which is usually behind it, leaving composers who listen to machine-generated pieces feeling unsettled by the lack of apparent meaning.

History

In the 1950s and the 1960s, music made by artificial intelligence was not fully original, but generated from templates that people had already defined and given to the AI, with this being known as rule-based systems. As time passed, computers became more powerful, which allowed machine learning and artificial neural networks to help in the music industry by giving AI large amounts of data to learn how music is made instead of predefined templates. By the early 2000s, more advancements in artificial intelligence had been made, with generative adversarial networks (GANs) and deep learning being used to help AI compose more original music that is more complex and varied than possible before. Notable AI-driven projects, such as OpenAI’s MuseNet and Google’s Magenta, have demonstrated AI’s ability to generate compositions that mimic various musical styles.

Timeline

Artificial intelligence finds its beginnings in music with the transcription problem: accurately recording a performance into musical notation as it is played. Père Engramelle's schematic of a "piano roll", a mode of automatically recording note timing and duration in a way which could be easily transcribed to proper musical notation by hand, was first implemented by German engineers J.F. Unger and J. Hohlfield in 1952.

In 1957, the ILLIAC I (Illinois Automatic Computer) produced the "Illiac Suite for String Quartet", a completely computer-generated piece of music. The computer was programmed to accomplish this by composer Leonard Isaacson and mathematician Lejaren Hiller.
In 1960, Russian researcher Rudolf Zaripov published worldwide first paper on algorithmic music composing using the Ural-1 computer.

In 1965, inventor Ray Kurzweil developed software capable of recognizing musical patterns and synthesizing new compositions from them. The computer first appeared on the quiz show '' I've Got a Secret'' that same year.

By 1983, Yamaha Corporation's Kansei Music System had gained momentum, and a paper was published on its development in 1989. The software utilized music information processing and artificial intelligence techniques to essentially solve the transcription problem for simpler melodies, although higher-level melodies and musical complexities are regarded even today as difficult deep-learning tasks, and near-perfect transcription is still a subject of research.

In 1997, an artificial intelligence program named Experiments in Musical Intelligence (EMI) appeared to outperform a human composer at the task of composing a piece of music to imitate the style of Bach. EMI would later become the basis for a more sophisticated algorithm called Emily Howell, named for its creator.

In 2002, the music research team at the Sony Computer Science Laboratory Paris, led by French composer and scientist François Pachet, designed the Continuator, an algorithm uniquely capable of resuming a composition after a live musician stopped.

Emily Howell would continue to make advancements in musical artificial intelligence, publishing its first album ''From Darkness, Light'' in 2009. Since then, many more pieces by artificial intelligence and various groups have been published.

In 2010, Iamus became the first AI to produce a fragment of original contemporary classical music, in its own style: "Iamus' Opus 1". Located at the Universidad de Malága (Malága University) in Spain, the computer can generate a fully original piece in a variety of musical styles. In August 2019, a large dataset consisting of 12,197 MIDI songs, each with their lyrics and melodies, was created to investigate the feasibility of neural melody generation from lyrics using a deep conditional LSTM-GAN method.

With progress in generative AI, models capable of creating complete musical compositions (including lyrics) from a simple text description have begun to emerge. Two notable web applications in this field are Suno AI, launched in December 2023, and Udio, which followed in April 2024.

Software applications

ChucK

Developed at Princeton University by Ge Wang and Perry Cook, ChucK is a text-based, cross-platform language. By extracting and classifying the theoretical techniques it finds in musical pieces, the software is able to synthesize entirely new pieces from the techniques it has learned. The technology is used by SLOrk (Stanford Laptop Orchestra) and PLOrk (Princeton Laptop Orchestra).

Jukedeck

Jukedeck was a website that let people use artificial intelligence to generate original, royalty-free music for use in videos. The team started building the music generation technology in 2010, formed a company around it in 2012, and launched the website publicly in 2015. The technology used was originally a rule-based algorithmic composition system, which was later replaced with artificial neural networks. The website was used to create over 1 million pieces of music, and brands that used it included Coca-Cola, Google, UKTV, and the Natural History Museum, London. In 2019, the company was acquired by ByteDance.

MorpheuS

MorpheuS is a research project by Dorien Herremans and Elaine Chew at Queen Mary University of London, funded by a Marie Skłodowská-Curie EU project. The system uses an optimization approach based on a variable neighborhood search algorithm to morph existing template pieces into novel pieces with a set level of tonal tension that changes dynamically throughout the piece. This optimization approach allows for the integration of a pattern detection technique in order to enforce long term structure and recurring themes in the generated music. Pieces composed by MorpheuS have been performed at concerts in both Stanford and London.

AIVA

Created in February 2016, in Luxembourg, AIVA is a program that produces soundtracks for any type of media. The algorithms behind AIVA are based on deep learning architectures AIVA has also been used to compose a Rock track called ''On the Edge'', as well as a pop tune ''Love Sick'' in collaboration with singer Taryn Southern, for the creation of her 2018 album "I am AI".

Google Magenta

Google's Magenta team has published several AI music applications and technical papers since their launch in 2016. In 2017 they released the NSynth algorithm and dataset, and an open source hardware musical instrument, designed to facilitate musicians in using the algorithm. The instrument was used by notable artists such as Grimes and YACHT in their albums. In 2018, they released a piano improvisation app called Piano Genie. This was later followed by Magenta Studio, a suite of 5 MIDI plugins that allow music producers to elaborate on existing music in their DAW. In 2023, their machine learning team published a technical paper on GitHub that described MusicLM, a private text-to-music generator which they'd developed.

Riffusion

Spike AI

Spike AI is an AI-based audio plug-in, developed by Spike Stent in collaboration with his son Joshua Stent and friend Henry Ramsey, that analyzes tracks and provides suggestions to increase clarity and other aspects during mixing. Communication is done by using a chatbot trained on Spike Stent's personal data. The plug-in integrates into digital audio workstation.

Musical applications

Artificial intelligence can potentially impact how producers create music by giving reiterations of a track that follow a prompt given by the creator. These prompts allow the AI to follow a certain style that the artist is trying to go for. AI has also been seen in musical analysis where it has been used for feature extraction, pattern recognition, and musical recommendations. New tools that are powered by artificial intelligence have been made to help aid in generating original music compositions, like AIVA (Artificial Intelligence Virtual Artist) and Udio. This is done by giving an AI model data of already-existing music and having it analyze the data using deep learning techniques to generate music in many different genres, such as classical music or electronic music.

Ethical and legal considerations

Several musicians such as Dua Lipa, Elton John, Nick Cave, Paul McCartney and Sting have criticized the use of AI in music and are encouraging the UK government to act on this matter.

Some artists have encouraged the use of AI in music such as Grimes.

While helpful in generating new music, many issues have come up since artificial intelligence has begun making music. Some major concerns include how the economy will be impacted with AI taking over music production, who truly owns music generated by AI, and a lower demand for human-made musical compositions. Some critics argue that AI diminishes the value of human creativity, while proponents see it as an augmentative tool that expands artistic possibilities rather than replacing human musicians.

Additionally, concerns have been raised about AI's potential to homogenize music. AI-driven models often generate compositions based on existing trends, which some fear could limit musical diversity. Addressing this concern, researchers are working on AI systems that incorporate more nuanced creative elements, allowing for greater stylistic variation.

Another major concern about artificial intelligence in music is copyright laws. Many questions have been asked about who owns AI generated music and productions, as today’s copyright laws require the work to be human-authorized in order to be granted copyright protection. One proposed solution is to create hybrid laws that recognize both the artificial intelligence that generated the creation and the humans that contributed to the creation.

In the United States, the current legal framework tends to apply traditional copyright laws to AI, despite its differences with the human creative process. However, music outputs solely generated by AI are not granted copyright protection. In the compendium of the U.S. Copyright Office Practices, the Copyright Office has stated that it would not grant copyrights to "works that lack human authorship" and "the Office will not register works produced by a machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human author." In February 2022, the Copyright Review Board rejected an application to copyright AI-generated artwork on the basis that it "lacked the required human authorship necessary to sustain a claim in copyright." The usage of copyrighted music in training AI has also been a topic of contention. One instance of this was seen when SACEM, a professional organization of songwriters, composers, and music publishers demanded that PozaLabs, an AI music generation startup refrain from utilizing any music affiliated with them for training models.

The situation in the European Union (EU) is similar to the US, because its legal framework also emphasizes the role of human involvement in a copyright-protected work.Bulayenko, Oleksandr; Quintais, João Pedro; Gervais, Daniel J.; Poort, Joost (February 28, 2022)
"AI Music Outputs: Challenges to the Copyright Legal Framework"
. ''reCreating Europe Report''. Retrieved 2024-04-03. According to the European Union Intellectual Property Office and the recent jurisprudence of the Court of Justice of the European Union, the originality criterion requires the work to be the author's own intellectual creation, reflecting the personality of the author evidenced by the creative choices made during its production, requires distinct level of human involvement. The reCreating Europe project, funded by the European Union's Horizon 2020 research and innovation program, delves into the challenges posed by AI-generated contents including music, suggesting legal certainty and balanced protection that encourages innovation while respecting copyright norms. The recognition of AIVA marks a significant departure from traditional views on authorship and copyrights in the realm of music composition, allowing AI artists capable of releasing music and earning royalties. This acceptance marks AIVA as a pioneering instance where an AI has been formally acknowledged within the music production.

The recent advancements in artificial intelligence made by groups such as Stability AI, OpenAI, and Google has incurred an enormous sum of copyright claims leveled against generative technology, including AI music. Should these lawsuits succeed, the machine learning models behind these technologies would have their datasets restricted to the public domain. Strides towards addressing ethical issues have been made as well, such as the collaboration between Sound Ethics(a company promoting ethical AI usage in the music industry) and UC Irvine, focusing on ethical frameworks and the responsible usage of AI.

Musical deepfakes

A more nascent development of AI in music is the application of audio deepfakes to cast the lyrics or musical style of a pre-existing song to the voice or style of another artist. This has raised many concerns regarding the legality of technology, as well as the ethics of employing it, particularly in the context of artistic identity. Furthermore, it has also raised the question of to whom the authorship of these works is attributed. As AI cannot hold authorship of its own, current speculation suggests that there will be no clear answer until further rulings are made regarding machine learning technologies as a whole. Most recently, preventative measures have started to be developed by Google and Universal Music group who have taken in royalties and credited attribution in order to allow producers to replicate the voices and styles of artists.

"Heart on My Sleeve"

In 2023, an artist known as ghostwriter977 created a musical deepfake called " Heart on My Sleeve" that cloned the voices of Drake and The Weeknd by inputting an assortment of vocal-only tracks from the respective artists into a deep-learning algorithm, creating an artificial model of the voices of each artist, to which this model could be mapped onto original reference vocals with original lyrics. The track was submitted for Grammy consideration for the best rap song and song of the year. It went viral and gained traction on TikTok and received a positive response from the audience, leading to its official release on Apple Music, Spotify, and YouTube in April 2023. Many believed the track was fully composed by an AI software, but the producer claimed the songwriting, production, and original vocals (pre-conversion) were still done by him. It would later be rescinded from any Grammy considerations due to it not following the guidelines necessary to be considered for a Grammy award. The track would end up being removed from all music platforms by Universal Music Group. The song was a watershed moment for AI voice cloning, and models have since been created for hundreds, if not thousands, of popular singers and rappers.

"Where That Came From"

In 2013, country music singer Randy Travis suffered a stroke which left him unable to sing. In the meantime, vocalist James Dupré toured on his behalf, singing his songs for him. Travis and longtime producer Kyle Lehning released a new song in May 2024 titled " Where That Came From", Travis's first new song since his stroke. The recording uses AI technology to re-create Travis's singing voice, having been composited from over 40 existing vocal recordings alongside those of Dupré.

Technical approaches

Artificial intelligence music encompasses a number of technical approaches used for music composition, analysis, classification, and suggestion. Techniques used are drawn from deep learning, machine learning, natural language processing, and signal processing. Current systems are able to compose entire musical compositions, parse affective content, accompany human players in real-time, and acquire patterns of user and context-dependent preferences.

Symbolic music composition

Symbolic music generation is the generation of music in discrete symbolic forms such as MIDI, where note and timing are precisely defined. Early systems employed rule-based systems and Markov models, but modern systems employ deep learning to a large extent. Recurrent Neural Networks (RNNs), and more precisely Long Short-Term Memory (LSTM) networks, have been employed in modeling temporal dependencies of musical sequences. They may be used to generate melodies, harmonies, and counterpoints in various musical genres.

Transformer models such as Music Transformer and MuseNet became more popular for symbolic generation due to their ability to model long-range dependencies and scalability. These models were employed to generate multi-instrument polyphonic music and stylistic imitations.

Audio-based music generation

This method generates music as raw audio waveforms instead of symbolic notation. DeepMind's WaveNet is an early example that uses autoregressive sampling to generate high-fidelity audio. Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs) are being used more and more in new audio texture synthesis and timbre combination of different instruments.

NSynth (Neural Synthesizer), a Google Magenta project, uses a WaveNet-like autoencoder to learn latent audio representations and thereby generate completely novel instrumental sounds.

Music information retrieval (MIR)

Music Information Retrieval (MIR) is the extraction of musically relevant information from audio recordings to be utilized in applications such as genre classification, instrument recognition, mood recognition, beat detection, and similarity estimation. CNNs on spectrogram features have been very accurate on these tasks. SVMs and k-Nearest Neighbors (k-NN) are also used for classification on features such as Mel-frequency cepstral coefficients (MFCCs).

Hybrid and interactive systems

Hybrid systems combine symbolic and sound-based methods to draw on their respective strengths. They can compose high-level symbolic compositions and synthesize them as natural sound. Interactive systems in real-time allow for AI to instantaneously respond to human input to support live performance. Reinforcement learning and rule-based agents tend to be utilized to allow for human–AI co-creation in improvisation contexts.

Affective computing and emotion-aware music systems

Affective computing techniques enable AI systems to classify or create music based on some affective content. The models use musical features such as tempo, mode, and timbre to classify or influence listener emotions. Deep learning models have been trained for classifying music based on affective content and even creating music intended to have affective impacts.

AI-based music recommendation systems

Music recommenders employ AI to suggest tracks to users based on what they have heard, their tastes, and information available in context. Collaborative filtering, content-based filtering, and hybrid filtering are most widely applied, deep learning being utilized for fine-tuning. Graph-based and matrix factorization methods are used within commercial systems like Spotify and YouTube Music to represent complex user-item relationships.

AI for automatic mixing and mastering

AI is also used in audio engineering automation such as mixing and mastering. Such systems level, equalize, pan, and compress to give well-balanced sound outputs. Software such as LANDR and iZotope Ozone utilize machine learning in emulating professional audio engineers' decisions.

Lyrics generation and songwriting aid

Natural language generation also applies to songwriting assistance and lyrics generation. Transformer language models like GPT-3 have also been proven to be able to generate stylistic and coherent lyrics from input prompts, themes, or feeling. There even exist AI programs that assist with rhyme scheme, syllable count, and poem form. .

Multimodal and cross-modal systems

Recent developments include multimodal AI systems that integrate music with other media, e.g., dance, video, and text. These can generate background scores in synchronization with video sequences or generate dance choreography from audio input. Cross-modal retrieval systems allow one to search for music using images, text, or gestures.

Cultural impact

The advent of AI music has caused heated cultural debates, especially its impacts on creativity, morality, and audience. As much as there have been praises about the democratization of music production, there have been fears raised about its impacts on producers, audience, and society in general.

Reactions and controversies

The most contentious application of AI music creation has been its misuse to produce offensive work. The music AI platforms have been used in several instances to produce songs with offensive lyrics that were racist, antisemitic, or contained violence and have tested moderation and accountability in generative AI platforms.[] The case has renewed argument about accountability in users and developers in producing moral outputs in generative models.

Aside from that, there have been several producers and artists denouncing the use of AI music due to threats to originality, handmade craftsmanship, and cultural authenticity. The music created by AIs lacks the emotional intelligence and lived life upon which human work relies, according to its critics. The concern comes in an era when there are steadily more songs made from AIs appearing on platforms and which others consider lowering human artistry. ">ref name=":1">

Musicians vs. consumers

Interestingly enough, while professional musicians have been generally more dismissive about using AI in music production, the general consumer or listener has been receptive or neutral to the idea. Surveys have found that in a commercial context, the average consumer often doesn't know or even care whether they hear music made by human beings or AI and that a high percentage says that it doesn't affect their enjoyment. ">ref name=":1" />The contrast between artist sentiment and consumer sentiment may hold far-reaching consequences in terms of the future economics within the music industry and the worth assigned to human creativity.

Public perception and general perception

The cultural value placed on AI music is similarly related to overall popular perceptions regarding generative AI. How generative AI-produced work—whether music or writing—is received in human terms has been found to be dependent upon such factors as emotional meaning and authenticity.[] As long as the output from AI proves persuasive and engaging, audiences may in some cases be willing to accept music whose author is not a human being, with the potential to reshape conventions regarding creators and creativity.

Future directions

The field of music and artificial intelligence is still evolving. Some of the possible key future directions for advancement include advancements in generation models, changes in how humans and AI collaborate musically, and the development of legal and ethical frameworks to address the technology's impact.

Advancements in generation models

Future research and development is expected to move beyond established techniques such as Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs). More recent architectures such as diffusion models and transformer based networks are showing promise for generating more complex, nuanced, and stylistically coherent music. These models may lead to higher quality audio generation and better long term structure in music compositions.

Human-AI collaboration

Besides the act of generation itself, a significant future direction of interest involves deepening the collaboration between human musicians and AI. Developments are increasingly focused on understanding the way these collaborations can occur, and how they can be facilitated to be ethically sound. This involves studying musicians’ perceptions and experiences with AI tools to inform the design of future systems.
Research actively explores these collaborative models in different domains. For instance, studies investigate how AI can be co-designed with professionals such as music therapists to act as supportive partners in complex creative and therapeutic processes, showing a trend towards developing AI not just as an output tool, but as an integrated component designed to augment human skills.

Regulatory changes and ethical considerations

As AI generated music becomes more capable and widespread, legal and ethical frameworks worldwide are expected to continue adapting. Current policy discussions have been focusing on copyright ownership, the use of AI to mimic artists (deepfakes), and fair compensation for artists. Recent legislative efforts and debates, such as those concerning AI safety and regulation in places like California, show the challenges involved in balancing innovation with potential risks and societal impacts. Tracking these developments is crucial for understanding the future of AI in the music industry.Hight, Jewly (25 April 2024). "AI music isn't going away. Here are 4 big questions about what's next". ''NPR''. Retrieved March 30, 2025. https://www.npr.org/2024/04/25/1246928162/generative-ai-music-law-technology

See also

* Algorithmic composition
* Automatic content recognition
* Computational models of musical creativity
* Generative artificial intelligence
* Generative music
* List of music software
* Music information retrieval
*

References

Further reading

Understanding Music with AI: Perspectives on Music Cognition
. Edited by Mira Balaban, Kemal Ebcioglu, and Otto Laske. AAAI Press.
Proceedings of a Workshop held as part of AI-ED 93
World Conference on Artificial Intelligence in Education on Music Education: An Artificial Intelligence Approach
*
Artificial Intelligence - Intelligent Art? Human-Machine Interaction and Creative Practice.
(Digital Society - Digitale Gesellschaft). Edited by Voigts, Eckart, Robin Auer, Dietmar Elflein, Sebastian Kunas, Jan Röhnert, Christoph Seelinger Bielefeld: transcript,2024.

External links

The Music Informatics Research Group
* ttps://web.archive.org/web/20190330070355/http://www.leeds.ac.uk/icsrim/ Interdisciplinary Centre for Research in Musicbr>Mixdevil - Is AI Good Gor Music ProducersOpenDreamAI full track generation

{{DEFAULTSORT:Music And Artificial Intelligence
Artificial intelligence art
Cognitive musicology
Computer music