AI Music Generator - coursework PDF

Preview

Summary

coursework...

Description

AI Music Generator

Team Members: Nitesh Kumar (17K-3676) Ashish Kumar (16K-3735) Manish Kumar (17K-3683)

Abstract: This paper discusses the idea of Artificial intelligence creating music. Main focus is on how and using which technologies and methods have this been possible and where does it go from here. Till now there have been a lot of research and development in this field. Many companies have developed software like IBM Watson beat, Google Magenta’s NSynth Super, Jukedeck, Melodrive, and Amper Music which are already helping artists be more creative in composition of music. These software systems apply machine learning, deep learning factors like neural networks which relies on analyzing large amount of data. By learning from input it generates its own unique melodies. This paper also talks about the future endeavors that may change the whole music industry. We may see AI taking the job of song-writers by writing its own lyrics which makes sense and are grammatically correct. Whether a computer (AI) can completely oust the musicians is still a philosophical question but the wonders musicians can achieve with help of AI can be surprising.

Introduction: What is Artificial Intelligence and how does it go hand in hand with Music? By this time, I’m sure more than half of the world’s population is familiar with what artificial intelligence is or even if they’re not they must’ve seen its implementation and its placement in our daily lives. AIs applications are more common than we can imagine. Collegiate definition of AI is “AI is a branch of

computer science dealing with the simulation of intelligent behavior in computers” [1]. From playing chess and defeating the best human chess players to recently launching the first privately owned crewed spacecraft in earth’s orbit. The ideas of musical automata and automatic music machines are centuries old. Increasing number of developments pointed to a new musical concept: artificial musical intelligence (AMI) [2]. AI simply offers more than a set of new paradigms for music. AI works as a collaborator and as a tool with musicians. There have been a lot of development in AMI past decade and many wonderous events are yet to be unfolded by AMI. If we dig deeper the connection between AI and music then it dates back to 1950s when research focused on algorithmic music. This paper is divided into three sections first sections discusses about the past researches, second section discusses about current research and last section discusses about what’s coming next. Software and music composition tools that has been developed with time are also mentioned in the paper along with the current software systems that are being used by different companies to compose music.

Problem Statement: There have been a lot of innovation and development in the software system/ application which compose algorithmic music based on the data they’re trained on. The art of music was considered as the god-gifted, because not everybody was able to create music according to their own wish and mood. But AMI (artificial musical intelligence) made it possible for everyone. From adding personalized music in games to using it in commercials the apps and software system are widely used by everyone. There is still a lot of research being done in and many companies have started digging deep into this tech. The motivation behind this research was to gather past work, methods and tools, analyze all useful on-going work, and what’s next in one paper. Purpose of Study: The purpose of study lies in the information that is presented in this research paper. Study all the past researches/ scholarly articles regarding my research topic, extract useful information, and put all that in one paper. This research paper will help the future scientists/artists/writers to have an insight on all aspects of this field in one paper. The research paper also evaluates the work which has been done since the beginning of artificial intelligence and artificial musical intelligence in its every aspect. Study the software systems which are developed by different companies and are being widely used by artists and every one interested in creating unique music. It will give readers partial overview of what’s happening in this field and where is it directed to? How AI composes music? Most of the systems work by using deep learning networks, a type of AI that’s reliant on analyzing large amount of data. Basically, you feed the software tons of material, from dance hits to disco classics, which

it then analyzes to find patterns. It picks up on things like chords, tempo, length, and how notes relate to one another, learning from all the input so it can write its own melodies [3]. Literature Review: This field has emerged in levels. From artists creating music without any assistance from AI to artists using AI for sound processing but not for composition to artists using AI to assist them in composing music to the music being created by AI and performed by humans to the music created and performed by AI. Sub-headings below are the part of literature review: Start of Algorithmic Music/ Early years In this section we discuss the start of AI and Music. The first algorithmic composition of music dates back to mid-1950s after the field of ‘artificial intelligence’ was invented officially. Though at the start of AI, music and AI were both different fields and nobody could say one would create another in future. Computers back then were very expensive and too slow to process anything. Having computer was considered the most valuable asset and not everyone had computers. 









The year was 1957, a year after the introduction of ‘Artificial Intelligence’ at a conference at Dartmouth college, in Hanover, new hemisphere. Lejaren Hiller and Leonard Isaacson from the university of Illinois at Urbana-Champaign programmed ‘iliac suite for string quartet’ [4], the first work completely written by artificial intelligence. In 1960, Russian researcher R.Kh.Zaripov, published the first paper on algorithmic music composition using the ‘ural-1’ [5] computer. As the research went further in understanding music, the level of music intelligence emerged in generative modeling [6] of music. In 1975 N. Rowe from the MIT Experimental Music Studio developed a system for intelligent music perception that enables a musician to play freely on an acoustic keyboard while the machine infers a meter, its tempo, and note durations. In 1980, David Cope from the university of California, santa cruz developed EMI (Experiments in Musical Intelligence). The system was based on generative models to analyze existing music and create new pieces based on them. In 1987 Iannis Xenakis, a renowned avant-garde composer, used stochastic algorithms to generate new raw material for his compositions.

Apart from this above-mentioned work in AI and music, there is also some other work which is equally notable. Olson’s (1961) dedicated computer was able to compose new melodies related to previously fed ones, using Markov processes.

Now we’ll talk about the composition methods. The general AI is divided into two areas: -soft computing-based music composition methods

-symbolic AI based music composition methods Soft computing-based music composition is further divided into three categories  Heuristic composition methods  Evolutionary Algorithms  Dynamic Programming  Deep learning composition methods  Deep belief networks  Convolutional networks  Recurrent networks  Stochastic Composition methods  Markov Models  Generative Adversarial networks Symbolic AI based music composition methods is also further divided into sub categories:  Agent composition methods  Declarative programming composition methods  Grammar composition methods Now we won’t go deep into explaining all these methods because main focus of this research is to look into the already existing software systems that are catering the music industry and helping musicians be more creative by using those software systems. For your interest you can read more about composition methods in the research paper I’m referring to. Ever since the beginning of AI and algorithmic music the focus of scientists in research have shifted from sound processing to music analysis to composition to performance to curation and education. Models such as music transcription, structure analysis, instrument recognition, emotion recognition are foundation of intelligent music applications. Humans have been practicing free hand on developing application to compose music and now lyrics as well. In 90s David Bowie developed and application called ‘Verbasizer’ [7] which would take random material as input and create new combinations as lyrics in the form of output. Now there have been many software systems that create sweet melodies which are very soothing. Sony, in 2016 used a software called Flow machine to create melodies. Now there are artists who are creating music albums with the help of AI for example Tayrn Southern who co-produced her album ‘I AM AI’ . Her new song ‘Break Free’ was composed by AI and she is using Amper Music platform, defined below, to compose music for her new album. Now some worth mentioning software systems are discussed below to shed more light in this research and complete its purpose. Software Systems/Application: IBM Watson Beat: This project was initiated by Janani Mukandan. It was released by IBM on GitHub. The first album it released was named as ‘Space’. It creates music based on the input melodies and rhythms. The composed music may not be as creative and unique as you may expect but we can call it

computational creativity. Several projects have been developed with the help of Watson beat one example is of Richard Daskas composing music for Redbull F1 commercial [8]. Methods used by Watson beat: It uses two methods of machine learning to create its composition: reinforcement learning and Deep Belief Network (DBN). You can pass many parameters as input to the Watson beat such as time, signature, tempo, and mood. Understanding from the input it creates the next note. For rhythm it uses reinforcement learning. Diagram attached below []

For creating melodies, it using Dee Belief Network (DBN). Picture is attached below

If want to check Watson beat create music according to exactly how you’re feeling right now you can check on GitHub []. Every time you run it; it’ll give different outputs so try with same input midi ‘ini’ parameter file.

Amper Music: also called the leader of artificial musical intelligence announced its ‘amper score’ [] world’s first end-to-end music composition platform. It was founded in 2014 with a purpose to help anyone create their own music without any expertise. It is fast, creative and music quality is best. It is enabling beginners and artists to compose custom music in seconds. Since the music is composed by AI you may feel less emotionally connected to the music because there is that human factor missing from its creation. Video editors are using amper score to find suitable music for their videos. “It is easy for me to create personalized music for my video in less than five minutes with amper score which used to take more than two hours before. The creation process is smooth and the product is unique” said the Anna Green. The algorithm is fed with a huge amount of music samples from different music genres as data which it uses for training itself. Then it finds the useful patterns and identify the key components to understand the type of music user wants to generate. It is capable of composing endless streams of music which are not limited to any country, any culture, or any genre. To summarize amper music is a software which is helping artists and users around the world compose soulful music according to their moods. Amper score is the product of amper music with which we can create endless number of melodies.

AIVA Music: AIVA was founded by Pierre Barreau in 2016 which specializes in classical and symphonic composition. It was recognized by music society SACEM []. AIVA is an artificial intelligence which is used to write unique music. It is used for composing emotional music for videos, games, commercials and any type of entertainment content. It learns through the music composed by humans and then generates the melodies and soundtracks as output. It trains on the large data of music partitions composed by human artists such as Moozart, Beethoven, Bach, etc. Its first studio album ‘Genesis’ was released in November 2016 second album ’among the stars’ in 2018 []. Due to short time and to keep research precise we’ve explained most used software systems which are currently in use by many music companies and are also globally available for everyone to use. Apart from above mentioned apps/ software systems there are also many other software systems which are deployed in the industry for making work unique and creative. Some examples of other software system are mentioned below Melodrive, Jukedeck, Spotify’s creator technology research lab, Google magenta, Note Performer, Nsynth, LANDR, Sony flow machines etc. These are some known and recognized software systems which

are using AI to compose music. Apart from this there also some artists like Tayrn Southern, who have been working all time on their albums being composed by AI.

Conclusion: AI was created by humans but what it is capable of achieving now may never have been achieved by humans. In near future we’ll listen to the music which will be composed by AI. The idea that a code is generating music is still fascinating and at the same time perplexing. Performance measurement is also the main concern of AMI. New methods will emerge in the near future and more algorithms will be developed. Many software systems and application are being developed which will focus on lyrics. Music creation process should not be limited to only the experts and professionals of field everyone should get an opportunity to create music. These applications and software systems provide the sense of reward to users when they create their own melodies and listen to them. What possibly could be better than composing music which belongs to you. AMI will empower the future generation of composers and content creators. According to a McKinsey report 70% of companies will have adopted at least one AI technology by 2030.

Recommendation: The future research in this field should cover the lyrics part. Write algorithms which should generate lyrics for new songs and the lyrics must be grammatically correct and make sense to the listeners. Music composers and songwriters must not hold fear of AI replacing them. Following are two possible scenarios for you to consider: -Applications that offers personalized music you can buy. After the purchase AI will search for all your information available online and then analyze what type of music you would like to hear and then generate it accordingly. Thus, satisfying customers personally like never before. -Let’s say you buy a ticket for online AI music concert and you do this by connecting your social profile. AI then analyzes all the potential visitors and categories them in different groups according to their taste of music. Then it generates the most appropriate music for each group and perform on different stages for better customer satisfaction. Thus, offering unprecedented experience to the attendees in concert.

Reference List: [1] Jeremy freeman, Artificial Intelligence and Music, feb 2020. Accessed on June. 10, 2020. [online]. Available: https://medium.com/@jeremy.freeman_53491/artificial-intelligence-and-music-what-thefuture-holds-79005bba7e7d

[2] Kumba Sennar, Musical Artificial Intelligence, Feb 2019. Accessed on June. 11, 2020. [online]. Available: https://emerj.com/ai-sector-overviews/musical-artificial-intelligence-6-applications-of-ai-foraudio/ [3] Dani Deahl, How AI-Generated Music Is Changing The Way Hits Are Made, August 2018. Accessed on June. 11, 202. [online]. Avaiable: https://www.theverge.com/2018/8/31/17777008/artificial-intelligencetaryn-southern-amper-music [4] Chong Li, A retrospective of AI + Music, Sep 2019. Accessed on June 11, 2020. [online]. Available: https://blog.prototypr.io/a-retrospective-of-ai-music-95bfa9b38531 [5] “Ural computer” accessed on June.11, 2020. [online]. Available: https://en.wikipedia.org/wiki/Ural_(computer) [6] Jifeng Dai, Generative modeling of convolutional neural netwroks, April 2015. Accessed on June. 11, 2020. [online]. Available: https://arxiv.org/abs/1412.6296 [7] Matthew Braga, The Verbasizer, Jan 2016. Accessed on June. 12, 2020. [online]. Available: https://www.vice.com/en_us/article/xygxpn/the-verbasizer-was-david-bowies-1995-lyric-writing-macapp [8] Anna Chaney, The Watson beat: using machine learning to inspire musical creativity, jab 2018. Accessed on June. 12,2020. [online]. Available: https://medium.com/@anna_seg/the-watson-beatd7497406a202...