From Sounds to Music: Learning the Bohlen-Pierce Scale Psyche Loui - PowerPoint PPT Presentation

From Sounds to Music: Learning the Bohlen-Pierce Scale Psyche Loui Beth Israel Deaconess Medical Center Harvard Medical School Bohlen-Pierce Scale Symposium March 7, 2010

The world knows and loves music

Whence musical knowledge? Perspectives:  Developmental studies  Cross-cultural studies  Artificial system Bohlen-Pierce scale 

The tritave as a musical system Bohlen-Pierce 700 F = 220 * 3 n /13 3 : 5 : 7 600 frequency (Hz) 500 400 300 200 0 1 2 3 4 5 6 7 8 9 10 11 12 13 increments (n)

Composing in the Bohlen-Pierce scale F = 220 * 3 n /13 10 7 10 10 6 4 7 6 0 0 3 0 Krumhansl, 1987 Loui & Wessel, 2008

Composing melody from harmony – applying a finite-state grammar 10 7 10 10 6 4 7 6 0 0 3 0 Loui & Wessel, 2008

Composing melody from harmony – applying a finite-state grammar 10 7 10 10 6 4 7 6 0 0 3 0 Melody: 10  10  4  7  6  10 Loui & Wessel, 2008

Can we learn the B-P scale? General design of behavioral studies: PRE-TEST 1. assess baseline  EXPOSURE to melodies in one grammar 2. ~30 minutes  POST-TESTS 3. assess learning 

Learning a musical system: basic questions  Can we recognize old melodies?  2-AFC test of recognition  Can we generalize to new melodies?  2-AFC test of generalization  Can we learn to like new melodies?  Preference ratings

Double dissociation between grammar learning and preference change recognition generalization preference change 100% 1.2 90% 1 (familiar - unfamiliar) Difference in rating Percent Correct 80% 0.8 70% 0.6 60% 0.4 50% 0.2 40% 0 No. of melodies 5 10 15 400 40 27 No. of repetitions 100 1 Loui & Wessel, 2008 Loui, Wessel & Hudson Kam, in press.

Learning a new musical system: more questions  Can we learn to expect frequent tones?  Probe tone ratings test  Probe tone profiles reflect frequencies of compositions Krumhansl, 1990

Testing for expectation for frequencies Probe tone ratings test (Krumhansl, 1990)  Melody  tone  Task: rate how well the tone fits the melody  Scale of 1 through 7  Tests conducted both pre- and post- exposure

Pre-exposure probe tone ratings 7 1200 Frequency of exposure 6 1000 5 800 Rating Rating 4 600 Exposure 3 400 2 200 1 0 0 1 2 3 4 5 6 7 8 9 10 11 12 F = 220* 3 n /13 Probe tone Loui, Wessel & Hudson Kam, in press.

Post-exposure probe tone ratings 7 1200 Frequency of exposure 6 1000 5 800 Rating Rating 4 600 Exposure 3 400 2 200 1 0 0 1 2 3 4 5 6 7 8 9 10 11 12 Probe tone Loui, Wessel & Hudson Kam, in press.

Correlating ratings with exposure 1 0.9 0.8 0.7 Correlation (r) 0.6 0.5 0.4 0.3 0.2 0.1 0 Pre Post Exposure Loui, Wessel & Hudson Kam, in press.

Sounds give rise to implicit learning of music Can we observe implicit learning in real time? Music Sounds w ith Event-Related Potentials: Yes

Event-Related Potentials can measure brain activity – Western music Early Anterior Negativity - 5 m V Fza 1000ms 150-200ms + Late Negativity Deviant 500-550ms Standard Loui et al, 2005

Event-Related Potentials can measure brain activity in the Bohlen Pierce scale  Experiment design: Chord progressions: Standard 70%  Deviant 20%  Fadeout 10%   Amplitude change detection task Attending to auditory stimuli but not to harmony  Dissociating perception from decision-making  Loui, Wu, Wessel, & Knight, 2009.

ERP responses to Bohlen-Pierce scale Early Anterior Negativity AFz [µV] Late -2 Negativity 150 – 210ms 0 2 0 500 [ms] Deviant Standard 500 – 550ms ERPs for improbable chords in B-P scale elicit EAN and LN. Loui, Wu, Wessel, & Knight, 2009.

Effects driven by probability? Probability of exposure vs. surface features of stimulus Equal probability AFz [µV] -2 0 2 0 500 [ms] Loui, Wu, Wessel, & Knight, 2009.

Learning probability during exposure Fz [µV] Standard Early -2 EAN peak amplitude 0 -5 2 Standard Late amplitude ( m V) 0 500 [ms] -4 Fz -3 [µV] Early -2 -2 0 -1 Deviant Early 2 0 0 500 [ms] Early Late Fz Late [µV] -2 0 Deviant Late 2 0 500 [ms] Loui, Wu, Wessel, & Knight, 2009.

ERP amplitude reflects individual differences 3 2.5 2 EAN amplitude ( m V) 1.5 1 R = 0.75 0.5 0 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 -0.5 -1 Generalization (proportion correct) Loui, Wu, Wessel, & Knight, 2009.

Statistics of sounds give rise to musical knowledge - Music + Sounds 7 6 5 4 3 2 1 0 0 1 2 3 4 5 6 7 8 9 10 11 12 Statistics and acoustics constrain music by limiting what we can learn.

Sound spectrum constrains knowledge in music 7 1200 Frequency of exposure 6 1000 5 800 Rating 4 600 3 400 2 200 1 0 0 1 2 3 4 5 6 7 8 9 10 11 12 Probe tone

Sound spectrum constrains knowledge in music

Sound spectrum constrains knowledge in speech and language? Wordle.net

Acknowledgements Center for New Music & Audio  David Wessel  Technologies Erv Hafter  Auditory Perception Lab  Carla Hudson Kam  Language & Learning Lab  Bob Knight  Knight Lab  UC Berkeley Psychology  Marty Woldorff (Duke)  Carol Krumhansl  Research Assistants  (Cornell) Charles Li Elaine Wu Shaochen Wu Pearl Chen Judy Wang Young Lee