jazzomat & dig that lick
TRANSCRIPT
Jazzomat & Dig That LickSome Assorted Results
Martin Pfleiderer
Simon Dixon
Klaus Frieler
MIRAGE Symposium #1:
Computational Musicology
Oslo, 9th June 2021
Studies
Corpus Studies
• Performance
• Microtiming & swing (WJD)
• Sound & intonation analysis (WJD)
• Style & History
• Case studies (e.g., Trane ./. Miles) (WJD)
• Feature history of jazz improvisation (WJD)
• Style classification post-bop vs. be/hard bop (WJD)
• Solo dramaturgy (WJD)
• Walking bass (WJD)
• Patterns
• Anatomy of a Lick (WJD, DTL1000)
• Pattern Use (WJD, DTL1000)
• Pattern Transmission (WJD, DTL1000)
• Psychology of Improvisation
• Jazz line grammar (WJD)
• Mid-level analysis (WJD)
• Easy First (WJD)
• Analysis-by-synthesis solo generation
Experimental Studies
• Perception of virtuosity, musicality, and emotions in jazz
• Development of jazz improvisation skills
Technical Papers
• Database description
• Transcription theory, practice, and algorithms
• Flexible Quantiziation algorithm
• Software manual, tutorials etc.
1. PERFORMANCE
Swing Ratio
• Swing here means playing “uneven eighths”.
• ~15,000 Swing Triples form the WJD.
• Performers show one or two clusters of
eighths: Even and swung.
• Average swing ratio for swung eighths:
1.42:1.
• Large variations with styles and performers.
• Single „soft“ swung cluster in earlier styles.
• „Hard“ swing as stylistic device in later
styles.
Corcoran & Frieler, Music Perception, 2021
Intonation of jazz performers
“On” if pitch in 25 cents window
Average Non/N = .72
Overall tendency to play sharp
Abeßer et al, IEEE/ACM Transactions on Audio, Speech and Language Processing, 2017
2. STYLE & HISTORY
Miles vs. Trane
John Coltrane Miles Davis
Rhythm Mostly fast lines
(„Sheets of sound“)
Rhythmically diverse
Longer tones & pauses
Intervals More descending More ascending
Fewer tone repetitions More tone repetitions
Larger intervals Smaller intervals
More thirds & arpeggios Fewer thirds
Pitch Larger pitch range Smaller pitch range
Avoids thirds, more blue notes
Midlevel Units More lines, expressive &
fragments
More licks, melody & void
Patterns More patterns Fewer patterns
No common pattern vocabulary
How did solos change over time?
Example: Pitch ranges (ρ = 0.596, p <.001)
Jazz solos over time
• Time
– Solos became longer
– Rhythms became more uniform
– More fast notes
– More lines
• Pitch
– Fifths of chords decrease
– Tonally more complex
– More chromatics
• Intervals
– Wider intervals
– More diverse interval combinations
• Sound
– Tones more stable
– Articulation more diverse
– Intonation more “precise”
• Expression
– Expanding ambitus
– Increasing exhaustion of pitch space
– Heightened expressivity
– More diverse overall design of solos
→ Overall trend to higher complexity, diversity
and expressivity.
Frieler, Jazz @100. Darmstädter Beiträge zur Jazzforschung (2018)
3. PATTERNS, PATTERNS
Patterns in Jazz
• Licks and formulas (patterns) important for jazz improvisation.
• Patterns are short snippets (N-grams) in a set of sequences (intervals,
pitches etc).
• Usually subject to certain properties (e.g., length, document &
relative frequencies).
• Licks are repeated patterns of a recognizable musical gestalt.
Pattern Example
18-tone interval N-gram by Bob Berg on „Angles“
[-2, 1, 1, -2, 1, -1, -1, -1, -1, 2, 2, -4, 2, -1, -1, 4, -2]
Measure 30
Measure 108
Pattern coverage100 % coverage by
N-grams with N ≥ 4
occurring at least in
two different solos
50-75 % coverage by
N-grams with N ≥ 4
occurring at least 32
times in at least
4 different solos
Pattern coverage: Number of notes in a solo contained in at least one interval N-gram
Solo frequency
Corp
us fre
quency
Pattern Transmission (Saxophonists, DTL1000 + WJD + Omnibook)
4. PSYCHOLOGY OF IMPROVISATION
Beaty, Frieler et al., Journal Exp. Psych. Gen. (2021)
Easy First(Easier stuff occurs earlier in a phrase)
Values ≤ 0
for lines
First increase, then
saturation
Values ≥ 0
for licks
Value = 0 for
simulations
Conclusion
• Large, well-curated, and richly annotated databases have immense
potentials for musicological research.
• Scientific use of a database strongly dependent on its quality.
• Algorithms change, data stays.
• Bottle-neck in digital jazz research: transcriptions.
• MIR techniques allow generation of very large databases, while
integrating data from various sources.
• But: Further need to improve quality of transcriptions and
annotations in order to dig the full potential of jazz databases.
• Desideratum: Move beyond the monophonic case.
Thank you!
The Jazzomat Team: Martin Pfleiderer, Jakob Abeßer, Wolf-Georg Zaddach,
Friederike Bartel, Benjamin Burkhard, Martin Breternitz, Peter Heppner, Yvette
Kneisel, Benedikt Koch, Simon Meininger, Benjamin Napravnik, Albrecht Probst,
Franziska Risch, Lydia Schulz, Amelie Zimmermann, Alaa Zouiten, Klaus Frieler
The DTL Team: Frank Höger, Simon Dixon, Polina Proutskova, Tillman Weyde, Daniel
Wolff, Helene-Camille Crayencour, Dogac Basaran, Geoffroy Peeters, Krin
Gabbard, Andrew Vogel, Gabriel Solis, Lucas Henry, Olga Velichkina, Klaus Frieler
Acknowledgements
DTL was funded under the Trans-Atlantic Program Digging into Data Challenge with
the support of the UK Economic and Social Research Council (ES/R004005/1), the
French National Research Agency (ANR-16-DATA-0005), the German Research
Foundation (PF 669/9-1), and the US National Endowment for the Humanities (NEH-
HJ-253587-17).
The Jazzomat Project was funded by the German Research Foundation (DFG-PF
669/7-1)