What Is Computer Generated Music?

Computer generated music is music produced in whole or in substantial part through automated computational processes, without requiring a human performer to play each note in real time. The field encompasses two distinct but overlapping activities: algorithmic composition, in which a program applies rules, probabilities, or learned models to generate musical scores or sequences, and computer sound synthesis, in which digital signal processing techniques produce the audio waveforms that realize those sequences. Work in this area draws on music theory, digital signal processing, artificial intelligence, and psychoacoustics, and it has been an active research area since the mid-twentieth century.

The earliest concrete milestone was the Illiac Suite (1957), composed by Lejaren Hiller and Leonard Isaacson using an early computer to apply counterpoint rules drawn from historical treatises. Stanford University's Center for Computer Research in Music and Acoustics (CCRMA) has documented the field's lineage from these origins through modern AI-based systems.

Algorithmic Composition Methods

Algorithmic composition encompasses a range of computational strategies for generating musical material. Stochastic methods use probability distributions to select pitches, rhythms, and dynamics, with Iannis Xenakis's use of Gaussian and Poisson distributions in compositions of the early 1960s as a formative example. Rule-based systems encode music-theoretic constraints such as voice leading, cadential patterns, and harmonic progressions as formal grammars or constraint sets, allowing a program to generate syntactically correct music within a defined style. Markov chains model each musical event as a probabilistic function of prior events, producing sequences that exhibit local stylistic consistency without global planning. More recently, evolutionary algorithms have been applied to music generation, maintaining a population of candidate phrases and selecting among them according to fitness criteria derived from music-theoretic heuristics or listener ratings. A comprehensive survey published on arXiv covering AI methods in algorithmic composition catalogs these approaches and their relative strengths and limitations.

Sound Synthesis and Digital Signal Processing

Generating audio, as distinct from generating note sequences, requires signal processing techniques that model how instruments and voices produce sound. Additive synthesis constructs timbres by summing sinusoidal components, each at a specified frequency, amplitude, and phase, allowing complex tones to be built from fundamental physics. Subtractive synthesis applies filters to a spectrally rich source waveform to sculpt a desired timbre, forming the basis of most analog synthesizers and their digital descendants. Frequency modulation (FM) synthesis, introduced commercially in the Yamaha DX7 synthesizer in 1983, uses one oscillator to modulate the frequency of another, producing a wide range of timbres from few parameters. Wavetable synthesis stores short recorded waveform cycles from real instruments and plays them back at variable pitch, a technique central to most software instruments used in contemporary music production. The ACM Communications article on algorithmic composition discusses how synthesis technologies and compositional algorithms have been combined in integrated systems.

AI-Based Music Generation

Modern deep learning approaches have brought significant changes to computer music generation. Recurrent neural networks (RNNs) and long short-term memory (LSTM) architectures process musical sequences as temporal data, learning to predict likely next events from large corpora of existing music in MIDI or audio form. Google's Magenta project applied these techniques to both melody generation and audio synthesis, and its NSynth neural synthesizer uses a WaveNet-like autoencoder to learn latent representations of instrument sounds and interpolate between them. Transformer architectures, originally developed for natural language processing, have proven effective for music generation tasks requiring long-range coherence, and models trained at scale produce output that maintains structural form over multi-minute durations.

Applications

Computer generated music has applications in a wide range of contexts, including:

• Film and video game scoring, where adaptive music systems respond dynamically to scene events
• Interactive installations and generative art, where continuous algorithmic output creates ever-changing soundscapes
• Music therapy, where personalized audio environments are generated to suit individual patient needs
• Training data generation, where synthetic music augments datasets for music information retrieval research
• Consumer streaming, where AI composition tools allow users to create background music without licensing