Johan Sundberg ATTEMPTS to REPRODUCE a PIANISTS EXPRESSIVE TIMING - - PowerPoint PPT Presentation

Presented by Beran Pacaci Sundberg, Friberg & Bresin: Director Musices ISE 599: Spring 2004 1

ATTEMPTS to REPRODUCE a PIANIST’S EXPRESSIVE TIMING with DIRECTOR MUSICES PERFORMANCE RULES

by JOHAN SUNDBERG, ANDERS FRIBERG and ROBERTO BRESIN

• BERAN PACACI -

SPRING 2004

Johan Sundberg

• Ph.D. in musicology at Uppsala

University.

• was the head of the music acoustics

research group from 1970 to 2001.

• a music performer as a singer.
• Current research areas include

– Singing Voice – Rules for Musical Performance – Breathing and Phonation

• http://www.speech.kth.se/~pjohan

Anders Friberg

• M.S. in Applied Physics. Ph.D. in ?
• a music performer as a pianist.
• Current research areas include

– Random fluctuations of timing in music performance. – Swing ratios and ensemble timing in jazz. – Combining the Radio Baton and Director Musices. – Performance synthesis in Director Musices.

• http://www.speech.kth.se/~andersfr/

Roberto Bresin

• Ph.D. in Music Acoustics.
• Current research areas include

– Analysis and synthesis of emotional content in music. – Articulation strategies in expressive music performance. – Emotional melodies for mobile phones. – Synthesis of Emotional Expression in Music Performance.

• http://www.speech.kth.se/~roberto

Presented by Beran Pacaci Sundberg, Friberg & Bresin: Director Musices ISE 599: Spring 2004 2

KTH Music Acoustics Group

• runs in four main streams:

– singing (solo and choral) – stringed instruments – music performance – computer music composition

Music Performance

• is an important research area today.
• The deviation from nominal inter-onset-

interval (IOI).

• Every musician plays differently.

Music Performance Rules

The rules are divided into three main classes: ÿ Differentiation Rules: enhance the differences between tone categories. ÿ Grouping Rules: show what tones belong together. ÿ Emphasis Rules: emphasize unexpected notes. ÿ Ensemble Rules: synchronize the various voices in an ensemble.

Music Performance Rules

• Differentiation Rules

– Duration Contrast – Melodic Charge – High Sharp

• Grouping Rules

– Punctuation – Double Duration – Tuning – Phrase Arch – Inégales (or swing) – Ensemble Swing – Final Ritard – Harmonic Charge

Presented by Beran Pacaci Sundberg, Friberg & Bresin: Director Musices ISE 599: Spring 2004 3

Director Musices (DM)

• a software for Automatic Music Performance.
• a rule system for Musical Performance.
• implementing all previously defined rules.
• features includes polyphony, midi I/O,

performance variable graphs and user rule definition.

Director Musices (DM)

• DM rules contain two elements:

ÿ Context ÿ Quantity

Director Musices (DM) Research Strategy

• Analysis-by-synthesis
• Analysis-by-measurement

Presented by Beran Pacaci Sundberg, Friberg & Bresin: Director Musices ISE 599: Spring 2004 4

Experiment Outline

• Comparison between the real pianist’s

performance with different performances produced by the DM system.

• MIDI data consists of score time in beats,

MIDI note numbers, IOI, sounded duration and dynamics in MIDI velocity.

Experiment Outline

• Initial Run: Each rule is tested one by one.

Duration Contrast, Melodic Charge, Punctuation, Double Duration, Phrase Arch, Harmonic Charge, Faster-uphill and Leap-tone-duration.

• Second Run: Combinations of rules are
• tested. First Phrase Arch Rule, then

combining with Harmonic Charge and Duration Contrast.

Limitation of Correlation

• The sign of the overall deviations is

considered, not their quantity.

• The correlation is highly sensitive to

extreme values.

• The correlation is much more sensitive to

the agreement for single notes when the number of notes compared is small as compared to when it is large.

Conclusion

• The Phrase Arch rule is the winner in the

experimented performance.

• Rule combination has to change between

sections.

• Time varying rule palettes would be a

worthwhile target.

• DM limitation: The general applicability of

the rule combination.

Presented by Beran Pacaci Sundberg, Friberg & Bresin: Director Musices ISE 599: Spring 2004 5

References

• Ramon Lopez de Mantaras & Josep Lluis Arcos (2002).

AI and Music From Composition to Expressive Performance.

• Efstathios Stamatatos & Gerhard Widmer (2002). Music

Performer Recognition Using an Ensemble of Simple Classifiers.

• The Science of Music Performance

http://www.speech.kth.se/music/performance/