Blog

Harmonizing Art and Technology

Artificial intelligence (AI) is the ability of machines to perform tasks that normally require human intelligence, such as learning, reasoning, and creativity. AI has been making remarkable progress in various fields, such as computer vision, natural language processing, and gaming. But what about music, especially classical music, which is often considered the pinnacle of human artistic expression?

Classical music is a genre of music that originated in the Western culture and is based on complex musical structures, forms, and styles. The composers use musical elements such as harmony, melody, rhythm, texture, and timbre to create musical pieces that convey emotions, ideas, and stories. Classical music is also performed by orchestras, which consist of different groups of instruments that play together in coordination.

AI has been influencing classical music in two main ways: composition and orchestration. Composition is the process of creating new musical pieces from scratch or based on existing musical material. Orchestration is the process of arranging a musical piece for a specific set of instruments or voices. Both processes require a high level of musical knowledge, creativity, and skill.

AI can assist human composers and orchestrators by providing them with new tools, techniques, and inspiration. AI can also generate original classical music pieces by itself, without any human intervention. This article delves into AI systems for classical music, exploring their advantages and challenges.

AI for Classical Music Composition

AI systems for classical music composition can be divided into two categories: rule-based and data-driven. Rule-based systems use predefined rules and algorithms to generate music according to certain musical principles and constraints. Data-driven systems use machine learning techniques to learn from large datasets of existing musical pieces and generate music based on the learned patterns and probabilities.

Rule-based Systems

Rule-based systems are based on the idea that music can be formalized as a set of rules and logic. For example, one can define rules for harmony, melody, rhythm, form, style, etc., and use them to generate music that follows these rules. Rule-based systems can also use mathematical models and algorithms to create music based on certain patterns or structures.

EMI (Experiments in Musical Intelligence), developed by David Cope in the 1980s, is an early rule-based system for classical music composition. The system can analyze existing musical pieces by famous composers such as Bach, Mozart, Beethoven, etc., and extract their musical features and rules. EMI can then use these features and rules to generate new musical pieces in the same style as the original composer.

Another example of rule-based systems for classical music composition is WolframTones, developed by Stephen Wolfram in 2005. WolframTones uses cellular automata, a type of mathematical model that simulates the behavior of complex systems using simple rules. WolframTones can generate musical pieces based on different cellular automata rules and parameters. The user can choose from various musical styles, such as classical, jazz, rock, etc., and customize the instruments, tempo, pitch range, etc.

Data-driven Systems

Data-driven systems are based on the idea that music can be learned from data. The systems use machine learning techniques, such as neural networks, to learn from large datasets of existing musical pieces and generate music based on the learned patterns and probabilities. Data-driven systems do not rely on predefined rules or algorithms but rather learn from the data itself.

One of the most recent examples of data-driven systems for classical music composition is MuseNet3, developed by OpenAI in 2019. MuseNet is a deep neural network that can generate 4-minute musical compositions with 10 different instruments and can combine styles from country to Mozart to the Beatles. MuseNet was not explicitly programmed with any musical knowledge but rather discovered patterns of harmony, rhythm, and style by learning to predict the next token in hundreds of thousands of MIDI files.

Another example of data-driven systems for classical music composition is AIVA (Artificial Intelligence Virtual Artist)4, developed by AIVA Technologies in 2016. AIVA is a deep neural network that can compose original classical music pieces for various purposes, such as films, games, commercials, etc. AIVA was trained on over 30,000 classical music scores by composers such as Bach, Mozart, Beethoven, etc., and can generate music in different genres, moods, tempos, etc.

AI for Classical Music Orchestration

AI systems for classical music orchestration can also be divided into two categories: rule-based and data-driven. Rule-based systems use predefined rules and algorithms to arrange a musical piece for a specific set of instruments or voices. Data-driven systems use machine learning on large musical datasets to generate new arrangements by learning patterns and probabilities.

Rule-based Systems

Rule-based systems are based on the idea that orchestration can be formalized as a set of rules and logic. For example, one can define rules for instrument ranges, timbres, roles, combinations, etc., and use them to generate orchestral arrangements that follow these rules. Rule-based systems can also use mathematical models and algorithms to create orchestral arrangements based on certain patterns or structures.

ORCHIDEE, an early rule-based system for classical music orchestration, emerged from IRCAM in the 1990s. ORCHIDEE is a software environment that allows composers to create and manipulate orchestral arrangements using various tools and techniques. The software can analyze existing musical pieces and extract their orchestral features and rules. ORCHIDEE can then use these features and rules to generate new orchestral arrangements or modify existing ones.

Another example of rule-based systems for classical music orchestration is Orchidea, developed by Carmine Emanuele Cella in 2018. Orchidea is a software tool that uses optimization algorithms to generate orchestral arrangements based on a given musical input. Orchidea can take a single melody or a polyphonic piece as input and generate an orchestral arrangement that maximizes the similarity with the input while satisfying certain constraints, such as instrument ranges, timbres, dynamics, etc.

Data-driven Systems

Data-driven systems are based on the idea that orchestration can be learned from data. They use machine learning techniques, such as neural networks, to learn from large datasets of existing orchestral arrangements and generate new arrangements based on the learned patterns and probabilities. The systems do not rely on predefined rules or algorithms but rather learn from the data itself.

DeepBach, a data-driven orchestration system for classical music, emerged in 2017, crafted by Gaëtan Hadjeres and François Pachet. DeepBach is a deep neural network that can generate four-part chorales in the style of Bach. The neural network, trained on 400+ Bach chorales, generates harmonically and contrapuntally coherent new chorales. DeepBach can also reharmonize existing melodies or modify existing chorales.

Another example of data-driven systems for classical music orchestration is FlowMachines, developed by Sony Computer Science Laboratories in 2016. FlowMachines is a software suite that uses machine learning techniques to generate music in various styles and genres. The software suite can take a musical input, such as a melody, a chord sequence, or a style, and generate a musical output, such as an arrangement, a variation, or a continuation. FlowMachines can also combine different musical inputs to create hybrid styles.

Advantages and Challenges of AI for Classical Music

AI has been influencing classical music in various ways, such as providing new tools, techniques, and inspiration for human composers and orchestrators; generating original classical music pieces by itself; and creating new musical styles and genres. AI has also been raising some advantages and challenges for classical music, such as:

Advantages:

• AI can enhance human creativity by offering new possibilities, suggestions, and feedback.
• AI can reduce human effort by automating tedious or repetitive tasks.
• AI can increase human productivity by speeding up the musical process.
• AI can expand human diversity by creating music in different styles, genres, cultures, etc.
• AI can improve human quality by generating music that is harmonious, coherent, expressive, etc.

Challenges:

• AI can pose ethical issues by creating music that is plagiarized, biased, harmful, etc.
• AI can raise aesthetic questions by creating music that is original, authentic, meaningful, etc.
• AI can cause social problems by creating music that is competitive, disruptive, alienating, etc.
• AI can face technical difficulties by creating music that is complex, dynamic, interactive, etc.
• AI can encounter musical limitations by creating music that is generic, predictable, boring, etc.

Conclusion

AI is influencing classical music composition and orchestration in various ways. AI can assist human composers and orchestrators by providing them with new tools, techniques, and inspiration. Artificial intelligence can also generate original classical music pieces by itself or create new musical styles and genres. AI has been developing different systems for classical music composition and orchestration based on rule-based or data-driven approaches. AI has also been raising some advantages and challenges for classical music in terms of ethics, aesthetics, social impact, technical feasibility, and musical quality.

FAQs: Classical Music AI