1. The Structure, Perception and Generation of Musical Patterns Nyssim Lefford Sonic Studio, Interactive Institute How do musical creators perceive structure in the patterns they create as they generate them? (focusing predominantly on rhythm) Three Experiments Primary Objectives: Tie a generative decision - a musical pattern - to some measure of the composer’s perceptions Influence of perceptions and constraints on generative process Strategy Preferences Similarity

2. Composition Games: Well-bounded: Constraints and Goals Comparable Results - Different responses to the same game - Responses to games with varying constraints Previously used in Design Research (Habraken and Gross) Simple, sparse and rhythmic music games Get at the fundamentals of the generative process Subject Population in all experiments: Subjects all play and instrument or sing. They varied in age, gender, musical education and experience, instrument, and genre preference.

3. Experiment 1: Objectives: Learn to design games and select meaningful constraints Will similar constraints yield similar patterns? Similar strategies? Subjects make simple, rhythmic patterns click and drag sound samples numbered boxes without notation or waveform limited duration of silence Experiments 2 and 3: Preference and Similarity Do creators share similar ways of perceiving structure? Results: Patterns varied Similar strategies for describing structure – External – Internal

4. Preference and Similarity The “Nine” Rhythms 1  --  -- 2  --  -  - 3  --  --  4  -  -  -- 5  -  -  -  - 6  -  -  --  7  --  -- 8  --  -  - 9  --  --  time Baseline: perceptual tests to assess the population and contrast to the patterns generated

5. Preference Baseline Test Antecedent 1 Consequence 3 Consequence 7 Consequence 5 High  --  --  --  -- Low  --  --  --  --   --  --  -  -  --  Poor Predictors: age, gender, education and experience, instrument, genre preference Trends in results: Some pairs preferred by a significant percentage Groups who preferred one pair and also disliked other pairs Preference is highly variable …

6. Similarity Baseline Test 0.25.5.751.01.251.51.75 Pattern 1  --  -- Pattern 4  -  -  -- Pattern 7  --  -- Results: Multidimensional Scaling Patterns 1, 2 and 3 Patterns 4, 5 and 6 Patterns 7, 8 and 9 Subject 1 First two attacks Regularity and density Similar? Rate 0 … 10 Pattern 2 Pattern 1

7. Similarity Baseline Results Type I Type II Type III Found: Three different styles of perceiving of similarity Poor Predictors: age, gender, education and experience, instrument, genre preference Style 2 Style 3 Style 1

8. Nine Consequences Experiment 2: Make an Antecedent Game Make an Antecedent then Rank consequences – Preference and Context Make the least preferred consequence sound best Pattern 1 Pattern 2 Pattern 3 Pattern 4 Pattern 5 Pattern 6 Pattern 7 Pattern 8 Pattern 9 First make the antecedent Then rank the consequences

9. Subject 2 antecedent consequence - - - - antecedentconsequence - - - - Preference Ranked Pairs Representation of preferences in context But results not yet tied to baseline results Experiment 3… Most Preferred Consequences Least Preferred Consequences

10. Experiment 3: Chaining Game (10 subjects) Pattern 9 Pattern 6Pattern 1Pattern 5Pattern 2 Pattern 4Pattern 2Pattern 8Pattern 1 Pattern 2 Pattern 3 Pattern 4 Pattern 5 Pattern 6 Pattern 7 Pattern 8 Pattern 9 similar Task: Created a chain using 4 of the 9 rhythms Could use a rhythm more than once 4 pairs of chains Each chain in the pair started with a different rhythm but were perceived as similar Pairs: regular, irregular, dense and sparse Phrases or rhythmic groupings

11. Analysis of Experiment 3 Pattern 9Pattern 5Pattern 3Pattern 8 Chains plotted on MDS plots Similar ways of crossing similarity space Patterns coupled to similarity perceptions

12. Out of 80 chains: 41 Asymmetrical chains 27 Repetitive chains 12 Circular chains New Phrase Groupings: 39 Two phrases 10 Three phrases 15 Four phrases 16 other Similarity baseline/MDS plots (poor predictor) High-level criterion for structuring chains Chains: Asymmetrical eg. 6278 Circular eg. 9179 Repetitive eg. 9898

13. Conclusions Future work Add musical dimensions Beyond preference and similarity More complex games What did we learn from the experiments? Compared creator’s perception of similarity to the patterns they created 3 Styles of perceiving similarity Tracked individual creators’ preferences as they generated patterns a representation of a generative decision mapped to similarity Goals: Understanding how the creator perceives structure

14. Acknowledgements This research was conducted at the Media Lab at the Massachusetts Institute of Technology Nyssim Lefford Sonic Studio, Interactive Institute nyssim.lefford@tii.se The Structure, Perception and Generation of Musical Patterns