Who Am I?

Who Am I?

• I'm omar delarosa. 👋

Who Am I?

• I'm omar delarosa. 👋

• I'm a Senior Engineer at Spotify. 🎶

Who Am I?

• I'm omar delarosa. 👋

• I'm a Senior Engineer at Spotify. 🎶

• I play music in my spare time. 🎸 🎹

Who Am I?

• I'm omar delarosa. 👋

• I'm a Senior Engineer at Spotify. 🎶

• I play music in my spare time. 🎸 🎹

• I studied (almost) a minor's-worth of music theory courses. 👶🎼

Who Am I?

• I'm omar delarosa. 👋

• I'm a Senior Engineer at Spotify. 🎶

• I play music in my spare time. 🎸 🎹

• I studied (almost) a minor's-worth of music theory courses. 👶🎼

• I studied a master's-worth of computer science courses. 🤖🎓

Who Am I?

• I'm omar delarosa. 👋

• I'm a Senior Engineer at Spotify. 🎶

• I play music in my spare time. 🎸 🎹

• I studied (almost) a minor's-worth of music theory courses. 👶🎼

• I studied a master's-worth of computer science courses. 🤖🎓

• I put music on Instagram and YouTube sometimes.

Rhythm

• duration - how long a tone lasts

• beat - a single unit of rhythm

• measure - a regularly spaced group of beats

Rhythm & Fractions

• Durations are all described as fractions of a measure

Rhythm & Fractions

• Durations are all described as fractions of a measure

• 1/4 Note

Rhythm & Fractions

• Durations are all described as fractions of a measure

• 1/4 Note

• 1/8 Note

Rhythm & Fractions

• Durations are all described as fractions of a measure

• 1/4 Note

• 1/8 Note

• 1/16 Note

Fun With Fractions

• Not all are multiples of 2.

Fun With Fractions

• Not all are multiples of 2.

• Some interesting things happen when you mix up durations where the denominator of the fraction is a multiple of 3.

Fun With Fractions

• Not all are multiples of 2.

• Some interesting things happen when you mix up durations where the denominator of the fraction is a multiple of 3.

• This is common in generes such as hip hop.

Markov Chains (With Benefits)

Markov Chains (With Benefits)

• Markov chain transition matrices can be learned from a corpus text (or a MIDI file)

Markov Chains (With Benefits)

• Markov chain transition matrices can be learned from a corpus text (or a MIDI file)

• Data can be represented and stored easily as structured data formats such as JSON

Markov Chains (With Benefits)

• Markov chain transition matrices can be learned from a corpus text (or a MIDI file)

• Data can be represented and stored easily as structured data formats such as JSON

• Can extend to more complex, multi-state transition matrices, but this is outside the scope of this prez.

Rhythm as Code

Beat Grids

• Centered around 1/16th note ticks

• Can be difficult to "escape the grid" with durations < 1/16.

• Also tough to use subdivisions that are multiples of 3.

Rhythm and Computation

Beat Grids as Code

// 16-element arrays can represent rhythm patterns, but are tough to read.
const kicks   = [1,0,1,0,  0,0,1,0,  0,0,1,0,  0,0,1,0];
const snares  = [0,0,0,0,  1,0,0,0,  0,0,0,0,  1,0,0,0];
const hats    = [1,1,1,1,  1,1,1,1,  1,1,1,1,  1,1,1,1];
const cowbell = [0,0,0,0,  0,0,0,0,  0,0,0,0,  1,0,1,0];

Rhythm Notation

Beat Grids as "Words"

// Easier to read, CPU-trivial preprocessing
const kicks   = fmt('1010 0010 0010 0010');
const snares  = fmt('0000 1000 0000 1000');
const hats    = fmt('1111 1111 1111 1111');
const cowbell = fmt('0000 0000 0000 01010');

Rhythm Notation

Beat Grids as Lists of Words (a Language?)

// A list of patterns
const kick_patterns = [
    fmt("1010 0010 0010 0010"),
    fmt("1001 0001 0101 0010"),
    fmt("1000 0101 0100 0010"),
    fmt("1000 0010 0000 0100"),
];

Generative Beats

We could make two pattern sets

const kick_patterns = [
    fmt("1010 0010 0010 0010"),
    fmt("1001 0001 0101 0010"),
    fmt("1000 0101 0100 0010"),
    fmt("1000 0010 0000 0100"),
];
const snare_patterns = [
    fmt("0000 1000 0000 1000"),
    fmt("0010 1000 0000 1010"),
    fmt("0000 1000 0010 1000"),
];

Generative Beats

And randomly combine them

const kicks_sequence = [
    ..._.sample(kick_patterns),
    ..._.sample(kick_patterns),
    ..._.sample(kick_patterns),
    ..._.sample(kick_patterns),
];
const snare_sequence = [
    ..._.sample(snare_patterns),
    ..._.sample(snare_patterns),
    ..._.sample(snare_patterns),
    ..._.sample(snare_patterns),
];
playParallel(kicks_sequence, snare_sequence);

Generative Beats

Simple Markov Chain Implementation

class MarkovChain {
    constructor(obj = {}, states = [], initialState = 0) {
        this.graph = { ...obj };
        this.states = [...states];
        this.currentState = initialState;
    }
    set() {
        const newState = this.sample(this.graph[this.currentState]);
        this.currentState = newState;
    }
    next() {
        this.set();
        return this.states[this.currentState];
    }
    sample(list) {
        return list[Math.floor(list.length * Math.random())];
    }
}

Markov Chain of Notes

• Using an adjacency list instead of matrix (for readability, simplicity)

const NOTES = ["C", "D", "E", "F", "G", "A", "B"];
const G = {
    // Repeated notes represent higher probabilities
    0: [1, 1, 0, 3, 4, 5, 6], // 0 -> 1 is 2/7, the rest 1/7
    1: [0, 0, 2, 3], // 1 -> 0 is 1/2 the others 1/4
    2: [1, 3, 4],
    3: [4], // 3 -> 4 means state 4 always follows 3 or 1/1 probability
    4: [5],
    5: [5, 4, 1, 0],
    6: [2, 2, 2, 3, 3],
};
const mc = new MarkovChain(G, NOTES);

Markov Chain of Chords

const CHORDS = [
    "C maj", "D min", "E min", "F maj", "G maj", "A min", "B dim"];
// Favors I <-> IV, V -> I cadences
const G = {
    0: [3, 3, 3, 5],
    1: [2, 5],
    2: [3],
    3: [4, 4, 4, 1, 1],
    4: [0, 0, 0, 5],
    5: [1, 6],
    6: [4],
};
const mc = new MarkovChain(G, CHORDS);

Markov Chain of Rhythm Patterns

// Tuples of duration & number of beats
const HATS = [
    [16, 4],
    [12, 3],
    [24, 6],
    [32, 4],
    [48, 6],
    [64, 8],
];
// Favors steady 1/16 notes -- common in hip hop
const G = {
    0: [0, 0, 0, 0, 0, 0, 1, 2, 3, 4], // 0 -> 0 has 3/5 odds
    1: [0, 0, 0, 3],
    2: [0, 0, 0, 3],
    3: [2, 5],
    4: [2, 3, 4, 1],
    5: [3, 2, 4, 2, 2],
};

And So...

• Music is a graph.

And So...

• Music is a graph.

• Music sequences are walks along the edges of those graphs.

And So...

• Music is a graph.

• Music sequences are walks along the edges of those graphs.

• Markov chains are cool.

Hypercube Music Example

And arpeggiator sequence of nearest neighbor "chord nodes" on an 8-D hypercube:

And So...

• Beyond Markov Chains there are many other ways to compose algorithmically.

And So...

• Beyond Markov Chains there are many other ways to compose algorithmically.

• You can use cellular automata

And So...

• Beyond Markov Chains there are many other ways to compose algorithmically.

• You can use cellular automata

• Hypercubes

And So...

• Beyond Markov Chains there are many other ways to compose algorithmically.

• You can use cellular automata

• Hypercubes

• NLP tools

And So...

• Beyond Markov Chains there are many other ways to compose algorithmically.

• You can use cellular automata

• Hypercubes

• NLP tools

• LLMs and beyond

Links

• Slides:

  • codemusic.net
  • code-music.netlify.app
  • omardelarosa/code-music-demos

• Other Links:

  • Tone.js - full features JS code library
  • vis.js - dataviz library
  • Yotam Mann
  • Chris Wilson - "A Tale of Two Clocks"
  • Andrew Sorensen - "The Concert Programmer"
  • Sam Aaron - "Programming as Performance"