Composing Heavy Metal with Artificial Intelligence - GPT-2 Trained on Music - Demo

Tristan Behrens
21 Jan 202215:06

TLDRTristan, a musician, shares his innovative process of composing heavy metal music with AI. After training a deep neural network on thousands of songs, he integrates it into his workflow, enhancing his productivity. He demonstrates his web-based composer, which generates music in real-time, adjusting to the existing composition. The AI, based on GPT-2, not only predicts bars but also harmonizes with the overall piece, allowing Tristan to curate and refine the output into full songs, significantly reducing his composition time.

Takeaways

  • 😀 Tristan used an AI trained on heavy metal music to compose 36 songs across three albums in 2021.
  • 🎼 In 2022, Tristan is developing a new album and has moved away from Jupyter notebooks to a web-based tool for music composition.
  • 🛠️ The new web-based composer is built with HTML5, JavaScript, and CSS, and includes enhanced functionalities.
  • 🎹 The AI, based on GPT-2, is trained on 7000 heavy metal songs and can generate music that fits harmoniously with existing tracks.
  • 📈 Tristan can adjust the 'note density' to control the complexity and rhythm of the music generated by the AI.
  • 🥁 The process starts with composing drums, which can inspire the creation of guitar riffs and other instrumental parts.
  • 🎸 The AI can generate additional guitar parts that complement the existing music, even considering the harmony with drums and bass.
  • 🎹 Tristan demonstrated adding a piano accompaniment to the composition, showing the AI's versatility.
  • 🔄 The AI's predictions consider the context of the entire piece, not just individual bars.
  • 👨‍💻 Tristan curates the AI's suggestions, keeping what he likes and replacing what he doesn't with other ideas.
  • 🚀 The use of AI has significantly reduced the time it takes Tristan to create a song structure, from one working day to half that time.

Q & A

  • What is the main topic of the video?

    -The main topic of the video is the process of composing heavy metal music with artificial intelligence, specifically using a GPT-2 model trained on heavy metal music.

  • How many albums did Tristan release in 2021 using AI for music composition?

    -Tristan released three albums in 2021 using AI for music composition.

  • What is the total number of songs Tristan generated using AI in 2021?

    -Tristan generated a total of 36 songs using AI in 2021.

  • What is the name of Tristan's music project that can be found on streaming platforms?

    -Tristan's music project is called 'Hexagon Machine' and it can be found on Spotify and other streaming platforms.

  • What tool did Tristan initially use for composing music with AI?

    -Initially, Tristan used a tool based on Jupyter notebooks with a lot of widgets for composing music with AI.

  • Why did Tristan transition from using Jupyter notebooks to web technology?

    -Tristan transitioned from Jupyter notebooks to web technology because he outgrew the functionality available in Jupyter and wanted to develop a more tailored tool for his needs.

  • What is the name of the new composer tool Tristan developed using web technology?

    -The new composer tool Tristan developed using web technology does not have a specific name mentioned in the script, but it is described as being written in HTML5 with JavaScript and CSS.

  • How does the AI model assist Tristan in composing music?

    -The AI model assists Tristan by predicting bars of music that are harmonically fitting with the existing composition, taking into account all the other bars that already exist.

  • What is the role of the 'note density' setting in the composition process?

    -The 'note density' setting determines the amount of notes per bar in the composition, affecting the rhythmic complexity of the music.

  • How does Tristan incorporate various instruments into the composition?

    -Tristan starts with a basic rhythm, such as drums, and then adds other instruments like guitar and bass, adjusting and recomposing bars as needed to create a cohesive piece.

  • What is the significance of the piano addition in the composition?

    -The piano addition provides an extra layer of texture to the composition, showing the versatility of the AI model in generating different instrument parts.

  • How does Tristan refine the AI-generated music?

    -Tristan refines the AI-generated music by curating the ideas provided by the AI, keeping what he likes and replacing or adjusting what he doesn't to fit his vision for the song.

  • What is the AI model trained on and how does it assist in the composition process?

    -The AI model is trained on 7000 heavy metal songs and assists in the composition process by predicting and generating bars of music that fit with the existing composition, allowing Tristan to focus on the overall structure and arrangement.

Outlines

00:00

🎼 AI-Powered Music Composition Process

Tristan, a musician, discusses his method of composing music with artificial intelligence. He shares his history of using a deep learning network trained on heavy metal music for his 2021 albums, which generated 36 songs. In 2022, he moves away from Jupyter notebooks and develops a web-based composer using HTML5, JavaScript, and CSS, incorporating new functionalities. He demonstrates creating a new project, choosing the genre and note density, and starting composition with drums, which he finds inspiring. The process involves listening and adjusting the AI-generated music, aiming to create a harmonious piece.

05:01

🔧 Enhancing Music Composition with AI

Tristan continues his demonstration by adding guitar and bass to the drum sequence, using the AI to generate music that fits harmoniously with the existing composition. He explains that the AI, a GPT2 model trained on thousands of heavy metal songs, predicts not just individual bars but considers the entire composition for a cohesive sound. He experiments with different guitar parts, adjusting the note density to explore various musical ideas, and even adds a piano accompaniment to enrich the texture of the music. The process showcases the AI's ability to assist in the creative process, allowing Tristan to curate and refine the AI's suggestions.

10:01

🎹 Evolving the Composition with AI Assistance

Tristan further develops the music piece by experimenting with different musical elements and densities. He adjusts the drum patterns, introduces new guitar riffs, and refines the bass lines, all while relying on the AI to generate music that complements the existing composition. He emphasizes the control he has over the genre and musical style, and how the AI helps to speed up the creative process by providing immediate musical ideas. Tristan also discusses the potential for creating music in different heavy metal sub-genres, inviting viewer feedback to guide future compositions.

15:03

📘 Conclusion and Call to Action

In the final paragraph, Tristan wraps up his demonstration by summarizing the benefits of using AI for music composition. He highlights the efficiency gains from the AI's ability to generate harmonious musical ideas quickly, reducing the time spent on improvisation and theory. Tristan encourages viewers to like, subscribe, and comment with their ideas or preferences for heavy metal music, indicating a willingness to create more content based on audience engagement.

Mindmap

Keywords

💡Artificial Intelligence

Artificial Intelligence, often abbreviated as AI, refers to the simulation of human intelligence in machines that are programmed to think like humans and mimic their actions. In the context of the video, AI is used to compose music, particularly heavy metal music, by generating notes and melodies that fit within the genre's conventions.

💡Composing

Composing is the process of creating a piece of music by putting together individual notes, melodies, and harmonies to form a coherent musical work. In the video, the composer Tristan uses AI to assist in this process, allowing the machine to generate parts of the music that are later curated and refined by the human composer.

💡Deep Learning Network

A deep learning network is a type of artificial neural network with multiple layers used for learning and making decisions based on complex patterns. In the script, Tristan mentions that he has trained a deep learning network on heavy metal music, which is then used to generate musical ideas for his compositions.

💡Heavy Metal Music

Heavy metal music is a genre of rock music characterized by amplified distortion, extended guitar solos, emphatic beats, and overall loudness. The video focuses on using AI to compose within this genre, demonstrating how AI can be applied to a specific style of music.

💡Jupyter Notebooks

Jupyter Notebooks is an open-source web application that allows users to create and share documents containing live code, equations, visualizations, and narrative text. Tristan mentions using Jupyter Notebooks in his previous workflow for music composition with AI but has since transitioned to a web-based tool.

💡Web Technology

Web technology encompasses the different tools and platforms used to create and manage websites, including HTML5, JavaScript, and CSS. Tristan has developed a new music composition tool using web technology, which allows him to work more effectively with AI-generated music.

💡Note Density

Note density refers to the number of musical notes per measure or bar in a piece of music. In the video, Tristan adjusts the note density to control the complexity and rhythmic intensity of the AI-generated music.

💡Drum Sequence

A drum sequence is a series of drum beats arranged in a pattern to form the rhythmic foundation of a piece of music. In the script, Tristan starts the composition process by creating a drum sequence with the help of AI, which sets the tempo and rhythm for the rest of the composition.

💡Piano Roll Notation

Piano roll notation is a visual representation of music where notes are displayed on a grid, with vertical lines representing beats and horizontal lines representing pitches. Tristan mentions that being accustomed to this type of notation allows him to quickly understand and edit the AI-generated music.

💡GPT-2

GPT-2 is a type of language model developed by OpenAI that generates human-like text. In the context of the video, Tristan has trained GPT-2 on heavy metal music, allowing it to predict and generate musical bars that fit with the existing composition.

💡Musical Ideas

Musical ideas refer to the creative concepts or themes that form the basis of a piece of music. Tristan uses AI to generate these ideas, which he then curates and refines to create a full song structure with elements such as verses, choruses, and bridges.

Highlights

Tristan demonstrates the use of AI in composing heavy metal music with GPT-2.

In 2021, Tristan released three albums using a deep learning model trained on heavy metal music.

The AI provided notes for 36 songs across three albums.

AI integration into the composing pipeline is discussed.

Tristan's new tool development for 2022 is introduced.

A transition from Jupyter notebooks to web technology for music composition is highlighted.

The new composer is written in HTML5 with JavaScript and CSS.

Tristan creates a new project called 'Friday' in the composer.

The process of composing with heavy metal subgenre and note density settings is shown.

Starting the composition with drums and improvising with guitar is suggested.

A drum sequence is created and played for inspiration.

Tristan edits the drum sequence to create a rhythmic pattern.

Adding guitar to the composition and listening to the result is demonstrated.

The use of GPT2 to predict and harmonize bars of music is explained.

Tristan adds a second guitar layer to the composition.

Experimenting with note density to create different guitar parts is shown.

Adding a piano accompaniment to the heavy metal composition is demonstrated.

Tristan discusses the ability to edit and curate AI-generated music.

The efficiency of composing with AI assistance is highlighted, reducing the time to create a song structure.

Tristan invites viewers to subscribe and share ideas for future music genres.