I remember the first time I heard a Moog synthesizer.
It didn't sound like a piano, a guitar, an organ, or anything else I had heard before. It seemed almost impossible that an electronic device could create a sound that didn't already exist in the physical world. There were no strings vibrating, no reeds, no hammers striking metal strings. There was simply an electrical circuit producing a sound that could be shaped and controlled by a musician.
At the time, it seemed like the beginning of something completely new.
In one sense, it was. But in another sense, the Moog synthesizer was only the beginning of a technological journey that would eventually lead to something far more extraordinary: computers capable of generating entire musical performances from a description of what we want to hear.
Today, we can ask an artificial intelligence system for a song describing a particular style, mood, instrumentation, tempo, arrangement, and even vocal character. The computer can generate something that sounds remarkably like a musical performance.
The distance between that experience and the first Moog synthesizers is enormous.
But underneath all that technology is something surprisingly familiar.
Mathematics.
The mathematics of waves, frequencies, amplitude, sampling, Fourier analysis, and digital signal processing provides much of the foundation upon which modern computer music has been built. Even as we move from analog synthesizers to digital audio workstations and ultimately to generative AI, the underlying physics of sound has not changed.
And somewhere in the middle of all of this, Mr. Fourier deserves a little credit.
1. From Electronic Instruments to Artificial Intelligence
The history of computer-generated music is really a history of increasingly sophisticated ways of describing and manipulating sound.
The earliest electronic instruments demonstrated that electrical circuits could produce musical tones. Analog synthesizers then gave musicians control over the ingredients of those tones. Digital technology allowed those same concepts to be represented as numbers and manipulated by computers.
Sampling made it possible to record real instruments and play those recordings back under computer control. MIDI allowed computers to describe musical performances without actually containing the sound itself.
Then came the digital audio workstation.
Programs such as Cakewalk Sonar turned the computer into a recording studio, mixing console, synthesizer controller, sampler, effects processor, and mastering environment all at once.
And now we have generative AI systems such as Google's Lyria, which move the interface another step further. Instead of specifying every oscillator, filter, envelope, sample, and MIDI note, we can describe a musical result and allow a trained model to generate it.
The technology has changed dramatically.
The fundamental nature of sound has not.
2. A Very Short History of Electronic Music
The story begins before the Moog.
Early electronic instruments demonstrated that sound could be produced by manipulating electrical signals rather than mechanical instruments. The theremin, developed in the 1920s, is one of the best-known examples. It produced an electronic tone whose pitch and volume could be controlled without physically touching the instrument.
Tape-based electronic music followed, allowing composers to record, cut, splice, reverse, speed up, slow down, and otherwise manipulate recorded sound.
But the development of the modern synthesizer was a major turning point.
In the 1960s, Robert Moog and others developed modular synthesizers based on voltage-controlled oscillators, filters, amplifiers, and other electronic modules. For the first time, musicians had practical control over the individual components from which an electronic sound could be constructed.
The basic concept was remarkably powerful:
Generate a sound → filter it → shape it → control its volume.
That basic architecture remains recognizable in software synthesizers today.
The 1970s and 1980s brought increasingly sophisticated digital instruments. Frequency-modulation synthesis became famous through instruments such as the Yamaha DX7. Digital sampling became practical, allowing recorded sounds to become musical instruments.
Then came MIDI.
MIDI provided a standardized way for electronic instruments and computers to communicate musical instructions. A MIDI message could say, in effect:
Play this note, at this velocity, on this channel.
It did not have to contain the sound itself.
As computers became powerful enough to record and process digital audio in real time, the digital audio workstation emerged.
The studio itself became software.
And eventually, even the synthesizer became software.
3. What Is Sound?
Before we can understand how a computer creates music, we need to understand what it is creating.
Sound is a physical phenomenon. When a speaker produces a sound, its cone moves back and forth, creating variations in air pressure. Those pressure variations travel through the air as waves.
Your eardrum responds to those variations, and your auditory system interprets them as sound.
A microphone does essentially the opposite of a speaker. It converts variations in air pressure into an electrical signal.
That electrical signal can then be converted into numbers by an analog-to-digital converter.
Once the sound is represented digitally, a computer doesn't see a guitar, a piano, or a human voice.
It sees numbers.
A digital recording might therefore be thought of as a very long sequence:
0.0021, 0.0047, 0.0019, -0.0032, -0.0061...
Each number represents the amplitude of the audio signal at a particular instant in time.
At a sampling rate of 44,100 samples per second, a single second of mono audio contains 44,100 individual samples.
A three-minute song contains millions of them.
And yet those millions of numbers contain everything necessary to reproduce the original waveform with extraordinary accuracy.
4. The Mathematics of a Musical Note
The simplest possible sound is a sine wave.
A sine wave has three fundamental characteristics that are particularly important to us: frequency, amplitude, and phase.
Frequency determines how rapidly the waveform repeats. We perceive frequency primarily as pitch.
A frequency of 440 cycles per second, or 440 Hz, is the musical note A above middle C.
Amplitude describes the size of the waveform and is closely related to perceived loudness.
Phase describes where within its cycle the waveform begins.
A simple sine wave can be described mathematically as:
x(t) = A sin(2πft + φ)
That may look intimidating at first, but it is actually just a recipe.
A tells us how large the wave is. f tells us how fast it oscillates. t tells us where we are in time. φ tells us its phase.
Change those numbers and we change the sound.
The important thing is that a computer doesn't have to physically construct a sine wave. It can calculate one.
And if it can calculate a sine wave, it can calculate much more complicated sounds.
5. Fourier: The Man Who Helped Explain It All
This is where Joseph Fourier enters our story.
In the early nineteenth century, Fourier demonstrated a remarkable mathematical idea: complex periodic waveforms can be represented as combinations of simpler sinusoidal waves.
In other words, a complicated waveform can be understood as a collection of simpler frequencies.
This is one of the most important ideas in all of digital audio.
Imagine a musical instrument playing an A at 440 Hz.
The sound isn't normally just one 440 Hz sine wave. The instrument also produces additional frequencies called harmonics or overtones.
The fundamental might be 440 Hz. Additional components might occur at 880 Hz, 1320 Hz, 1760 Hz, and so on, with their relative strengths determined by the physical characteristics of the instrument.
Those combinations of frequencies are a major part of what gives an instrument its characteristic sound, or timbre.
A violin and a flute can play exactly the same musical note at exactly the same pitch, yet sound completely different.
Why? Because their frequency spectra are different.
Fourier analysis gives us a mathematical way to see those differences.
And this is where the story gets particularly interesting.
The waveform tells us what the sound looks like in time. Fourier analysis tells us what frequencies are inside it.
6. The Fourier Transform
The Fourier transform provides a mathematical conversion between two different ways of looking at a signal.
The time domain tells us how the amplitude changes over time.
The frequency domain tells us how much energy exists at different frequencies.
The Fast Fourier Transform, or FFT, is an extremely efficient computational method for performing this transformation.
Instead of looking at millions of audio samples and trying to understand the musical information hidden among them, we can analyze portions of the signal and determine which frequencies are present.
If we perform an FFT on a short slice of audio, we obtain a frequency spectrum. Now perform the same operation on the next slice. Then the next. Then the next.
Lay those frequency spectra out sequentially along a time axis and we obtain a time-frequency representation of the sound.
Conceptually, it is a three-dimensional map:
Time × Frequency × Magnitude
The X axis represents time. The Y axis represents frequency. The magnitude at each point tells us how strongly that frequency is present at that moment.
This is the mathematical foundation of the familiar spectrogram.
And this is a crucial concept for understanding computer music.
The FFT hasn't changed the sound. It has simply re-expressed the same sound in a mathematical form that makes its frequency structure visible.
7. Subtractive Synthesis: Starting With Too Much Sound
Now we can return to the Moog.
One of the fundamental techniques of analog synthesis is subtractive synthesis.
The basic idea is simple. Start with a waveform containing many frequencies and remove the frequencies you don't want.
A sawtooth wave, for example, contains a large number of harmonics. Feed that waveform into a filter and the filter can remove some of those frequencies.
A low-pass filter allows lower frequencies through while reducing higher frequencies. A high-pass filter does the opposite. A band-pass filter allows a range of frequencies to pass.
Resonance can emphasize frequencies near the filter's cutoff frequency, creating the characteristic sound associated with many analog synthesizers.
The synthesizer can then shape the sound over time using an envelope. The familiar ADSR envelope provides:
Attack → Decay → Sustain → Release
An oscillator creates the basic waveform. A filter shapes its frequency content. An amplifier controls its level. An envelope controls how those parameters change over time.
This may sound very different from modern AI music generation, but the underlying idea is remarkably similar to what computers still do today:
Represent sound mathematically, manipulate the representation, and turn the result back into sound.
8. Additive Synthesis: Building Sound From the Bottom Up
Additive synthesis approaches the same problem from the opposite direction.
Instead of starting with a complex waveform and removing frequencies, we start with individual sine waves and add them together.
One sine wave might provide the fundamental. Another provides the second harmonic. Another provides the third. And so on.
By controlling the amplitude of each component, we can construct increasingly complex timbres.
This is closely related to Fourier analysis. Fourier analysis asks: What frequencies make up this sound? Additive synthesis asks: What happens if I combine these frequencies to create a sound?
They are essentially two sides of the same mathematical idea.
The computer doesn't care whether we call the result a flute, a bell, or a synthesizer patch. It is adding and manipulating numerical waveforms.
9. Digital Synthesis
Once computers became fast enough, many forms of synthesis that were difficult or expensive in analog hardware became practical in software.
Digital synthesizers can generate waveforms mathematically. They can manipulate those waveforms with digital filters. They can calculate envelopes and modulation. They can use tables of precomputed waveforms. They can perform frequency modulation. They can combine synthesis techniques with recordings.
Modern software synthesizers can perform millions or billions of numerical calculations to produce a musical performance.
The underlying process is still signal processing. The difference is that instead of electrons flowing through dedicated analog circuits, a computer is performing mathematical operations on numbers.
10. Sampling: Turning Real Sounds Into Numbers
Synthesis isn't the only way a computer can create an instrument. We can also record one.
Digital sampling converts an analog waveform into a sequence of numerical measurements.
The sampling rate determines how frequently those measurements are taken. At 44.1 kHz, the computer takes 44,100 measurements every second.
Bit depth determines how precisely each measurement can be represented. The familiar 16-bit and 24-bit formats provide enormous numbers of possible amplitude values.
The mathematics of sampling also introduces an important limitation known as the Nyquist frequency. To accurately represent a frequency, the sampling rate must be more than twice that frequency. This is why a 44.1 kHz sampling rate can represent frequencies up to approximately 22.05 kHz.
Digital audio therefore isn't an approximation in the casual sense of the word. Properly sampled audio can reproduce the audible frequency range with extremely high accuracy.
11. Samplers and Virtual Instruments
A sampler takes recorded sounds and turns them into playable instruments.
A piano sampler might contain individual recordings of many different piano notes. It may contain multiple recordings of each note at different playing velocities. A sophisticated sampler may even contain different recordings for repeated notes so that the exact same sample isn't triggered every time.
The computer then decides which recording to play, how much to change its pitch, how loudly to play it, and how to combine it with other sounds.
This is another important step toward modern computer-generated music. The computer isn't necessarily generating every sound from mathematical oscillators. Sometimes it is reconstructing a performance from a library of recorded sounds.
12. MIDI: Instructions Instead of Sound
MIDI is frequently misunderstood.
MIDI is not digital audio. A MIDI sequence doesn't contain the sound of a piano note. It contains instructions that can tell an instrument to play a piano note.
A MIDI event can specify things such as which note to play, when to play it, when to stop it, how hard it was played, which instrument or channel should respond, and controller information such as modulation or sustain.
This distinction is important. A MIDI file might contain a beautiful piano performance, but without an instrument to interpret those instructions, there is no piano sound.
The MIDI information controls a synthesizer, sampler, or virtual instrument, which then generates the actual audio.
This gives us another useful conceptual distinction:
MIDI describes the performance. Audio contains the sound.
13. The Computer Becomes the Recording Studio
Eventually, computers became powerful enough to perform all of these tasks inside one application. The result was the Digital Audio Workstation, or DAW.
Cakewalk Sonar is an example. A modern DAW can simultaneously function as a multitrack recorder, a MIDI sequencer, a synthesizer controller, a sampler, a digital mixing console, an effects processor, an automation system, and a mastering environment.
A typical project might contain dozens or hundreds of tracks. Some tracks contain recorded audio. Others contain MIDI instructions controlling virtual instruments. Those instruments generate audio. The audio then passes through effects such as EQ, compression, reverb, delay, and limiting. Finally, all of those signals are combined into a stereo or multichannel mix.
And underneath the graphical interface, the DAW is performing an enormous number of mathematical operations.
14. What Happens to Audio Inside a DAW?
To understand the remarkable precision of a modern DAW, it helps to remember one thing:
Digital audio is numbers.
If the DAW needs to increase the volume of a sample, it performs multiplication. If it needs to mix two signals, it performs addition. If it needs to filter a signal, it performs a mathematical filtering operation. If it needs to apply automation, it changes parameters as a function of time. If it needs to pan a signal between left and right channels, it applies appropriate gain calculations to each channel.
Every plug-in is essentially a specialized mathematical processor.
This is where the extraordinary power of digital audio becomes apparent. The computer isn't physically manipulating sound while it is inside the DAW. It is manipulating mathematical representations of sound.
15. Why 32-Bit Floating Point Matters
Modern DAWs commonly use 32-bit floating-point arithmetic internally.
This doesn't mean that the final recording necessarily becomes a 32-bit file. It means that during processing, the computer has a very large numerical working range.
A floating-point representation provides tremendous headroom for internal calculations. A signal can temporarily exceed what would normally be considered 0 dBFS during internal processing without necessarily being permanently clipped at that point.
For example, suppose several tracks are summed together and the resulting internal signal becomes greater than 0 dBFS. In a floating-point environment, those values can still be represented. The engineer can subsequently reduce the level and bring the signal back into the normal range.
This is one of the great advantages of modern DAWs. The numbers being manipulated internally have enough precision and range that repeated operations such as gain changes, filtering, equalization, and effects can be performed with extremely small numerical errors.
The important distinction is that 32-bit floating point provides processing headroom and precision; it does not magically create additional information that wasn't present in the original recording.
16. Digital Filters and EQ
Remember the Moog filter? The same fundamental concept exists in the digital world.
A digital filter examines the samples in a signal and mathematically determines how much of various frequency components should be passed, reduced, or emphasized.
A low-pass filter reduces high frequencies. A high-pass filter reduces low frequencies. A band-pass filter isolates a range.
An EQ is essentially a collection of carefully designed filters that allow the engineer to control different portions of the frequency spectrum. A parametric EQ might provide controls for frequency, gain, and bandwidth or Q.
Move the frequency control and we change which portion of the spectrum is affected. Change the gain and we determine how much it is boosted or cut. Change Q and we determine how broad or narrow the affected region is.
Behind the knobs and graphs, the computer is calculating filter coefficients and applying them to the audio samples. What looks like dragging a curve with a mouse is actually mathematics being performed thousands of times per second.
17. Compression and Limiting
A compressor is also a mathematical device, although its operation is different from an EQ.
A compressor monitors the level of the incoming signal. When the signal exceeds a specified threshold, the compressor reduces its level according to a specified ratio.
Attack determines how quickly the compressor responds. Release determines how quickly it stops reducing the signal. Makeup gain can then be used to raise the resulting signal.
A limiter is essentially an extreme form of level control designed to prevent peaks from exceeding a specified ceiling.
Again, what appears to be a collection of knobs on a screen is actually a mathematical algorithm operating on the audio stream. The same principle applies to reverb, delay, chorus, distortion, saturation, pitch shifting, and countless other effects.
Modern music production is therefore an enormous collection of mathematical transformations applied to streams of numbers.
18. VST: The Common Language of Audio Plugins
The development of plugin standards made this enormously flexible system practical.
VST, or Virtual Studio Technology, provided a standardized way for software instruments and audio effects to communicate with a DAW.
A VST instrument can receive MIDI instructions from the DAW and produce audio. A VST effect can receive audio, process it, and return the modified audio to the DAW. The DAW can also automate the plugin's parameters.
This means that the same fundamental architecture can accommodate thousands of different instruments and effects. The signal path might look something like this:
MIDI → virtual instrument → audio → EQ → compressor → reverb → bus → master
Every stage is performing mathematical operations. The user sees instruments, knobs, sliders, meters, and graphical curves. The computer sees numbers and algorithms.
19. From Programmed Rules to Machine Learning
Up to this point, we have been telling the computer explicitly what to do.
We tell an oscillator to generate a waveform. We tell a filter what frequencies to remove. We tell a compressor what threshold and ratio to use. We tell a sampler which recording to play. We tell a MIDI sequencer which notes to play.
This is rule-based music generation.
Machine learning introduces a different approach. Instead of explicitly programming every rule describing what music should sound like, we can provide a computer with large amounts of musical information and allow a neural network to learn patterns and relationships within that information.
The computer can learn statistical relationships among things such as pitch, rhythm, harmony, timbre, instrumentation, arrangement, and musical structure.
The programmer doesn't necessarily have to write an individual rule saying:
"When you hear this combination of frequencies, it is a guitar."
The system can learn relationships from examples.
This represents a profound change in the way humans interact with computers to create music.
20. Fourier Is Still With Us
And this brings us back to Mr. Fourier.
It would be easy to think that once artificial intelligence entered the picture, all of the old mathematics of synthesis became obsolete.
It didn't.
The fundamental mathematical description of sound is still the same. A sound is still a physical waveform. That waveform can still be represented digitally as samples. Those samples can still be transformed into frequency-domain information using Fourier analysis. Successive FFTs can still reveal how the frequency content of a sound changes through time.
The same frequency-domain concepts that help us understand synthesizer waveforms also help us analyze recorded music. They are used in audio analysis, equalization, filtering, visualization, spectral processing, and many other areas of digital signal processing.
Modern AI music systems take this mathematical tradition into a new realm.
Google describes its Lyria models in terms of audio representations and temporal audio latents. The exact proprietary details of those internal representations aren't the important part for our purposes. What matters is that modern generative music systems continue to operate on mathematical representations of sound rather than somehow bypassing the mathematics of audio.
The remarkable thing is that the basic concepts introduced by Fourier nearly two hundred years ago remain relevant.
We have gone from electrical oscillators to digital signal processors to neural networks. But we are still describing sound in terms of things such as:
time, frequency, amplitude, phase, and their relationships.
So perhaps we should give Mr. Fourier a little credit. He was working on the mathematics long before anybody imagined a synthesizer, a DAW, or an artificial intelligence capable of writing a song.
21. Lyria and the New Interface to Music
The arrival of generative music models represents another major transition.
With a traditional synthesizer, the musician might specify: Oscillator → waveform → filter → envelope → amplifier.
With a sampler, the musician might specify: Instrument → sample → pitch → velocity → articulation.
With MIDI, the musician might specify: Note → timing → velocity → controller information.
With a DAW, the musician can combine all of these things and process them through a vast collection of digital signal processors.
With generative AI, the interface can become much more abstract.
Instead of specifying every individual component, the musician can describe the desired result. The system has learned relationships between musical concepts and sound from its training. The user supplies an intent. The model generates a musical result.
That doesn't mean the mathematics disappeared. It means the mathematics has moved deeper into the machine.
The knobs and patch cables of the Moog have been replaced by algorithms. The algorithms have increasingly been replaced—or supplemented—by learned models.
But at the end of the process, the result still has to become a waveform. And that waveform still has to move a speaker.
22. AI as a Collaborator, Not a Replacement
There is another way to look at generative music systems such as Lyria.
They don't have to replace the traditional music-production process. They can become part of it.
A musician can use an AI model as a creative collaborator—a source of ideas, musical possibilities, arrangements, performances, or individual parts that would otherwise take considerable time to create.
For example, an artist might begin with a song idea and lyrics, then use an AI music model to explore possible arrangements. One generation might suggest an interesting guitar part. Another might provide a rhythmic idea. A third might suggest a bass line, keyboard texture, vocal approach, or background vocal arrangement.
The artist doesn't have to accept the generated result as a finished song. Instead, the useful material can become stems—individual musical components that can be brought into a traditional DAW.
Once inside a DAW such as Cakewalk Sonar, those stems become just another collection of audio tracks. They can be edited, trimmed, time-aligned, pitch-corrected, EQ'd, compressed, reverb-treated, re-arranged, layered with traditionally recorded instruments, combined with MIDI instruments, replaced or supplemented with live performances, and automated and mixed.
The final song can therefore be a hybrid creation.
An artist might write the lyrics and melody, use AI to explore an arrangement, keep a generated guitar or percussion stem, replace the AI bass with a virtual instrument, record a live guitar part, add their own vocals, and then perform the entire mixing and mastering process inside the DAW.
In that workflow, the AI isn't the studio.
It is another instrument in the studio.
This is an important distinction.
The traditional synthesizer didn't eliminate musicians. It gave musicians another way to create sound. The sampler didn't eliminate instrumentalists. It gave producers another source of musical material. The DAW didn't eliminate the recording studio. It put much of the studio's equipment inside a computer.
Generative AI can be viewed in much the same way. It provides another creative instrument—one capable of generating musical ideas at a level of complexity that would be difficult to program manually.
The artist remains responsible for deciding what is useful, what isn't, what belongs in the song, and how all of the pieces should ultimately work together.
In this workflow, the creative process becomes a conversation:
Artist → AI → Artist → DAW → Artist → finished music
The AI provides possibilities. The artist provides judgment. The DAW provides control. And the final result can incorporate all three.
23. The Long Journey From Moog to Lyria
The evolution can therefore be viewed as a progression in how much of the musical construction we explicitly control.
Analog synthesizer — The musician directly controls electronic circuits.
Digital synthesizer — The musician controls mathematical algorithms.
Sampler — The musician controls recordings represented as digital data.
MIDI — The musician controls instructions describing a musical performance.
DAW — The musician controls an entire network of digital signal processors, instruments, recordings, and musical instructions.
Generative AI — The musician describes an intended musical result and a trained model generates the underlying musical and sonic structures.
The interface has changed enormously. The physics has not.
Sound is still vibration. Digital audio is still numbers representing that vibration. Fourier analysis still provides a powerful way of describing the frequency content of those numbers. Filters still manipulate frequency content. Envelopes still shape sound over time. Mixers still add signals together. Compressors still manipulate amplitude. Speakers still convert electrical signals into physical movement.
And our ears still receive the result as changing pressure waves.
24. The Complete Journey
We can now follow a musical idea through the entire modern production chain:
Musical idea → MIDI, performance, recording, or AI prompt → Synthesizer, sampler, virtual instrument, or generative model → Digital audio → Mathematical signal processing → EQ, filters, compression, limiting, effects → DAW mixing → Mastering → Digital-to-analog conversion → Amplifier → Speaker → Air pressure variations → Human hearing
The remarkable part is that every stage between the original musical idea and the final sound can now be represented, manipulated, and calculated by a computer.
That is what computer-generated music really is.
It isn't magic.
It is physics, mathematics, engineering, computer science, and music technology layered on top of one another for more than a century.
And the story that began with a strange electronic sound coming out of a Moog synthesizer has now reached a point where we can describe a musical idea in words and have an artificial intelligence construct the music for us.
The technology may be new. The interface may be astonishingly different.
But underneath it all, the waveform is still a waveform. The frequencies are still frequencies. The numbers are still numbers.
And somewhere in the mathematics, Mr. Fourier is still smiling.