Have a personal or library account? Click to login
Evaluating an Analysis-by-Synthesis Model for Jazz Improvisation Cover

Evaluating an Analysis-by-Synthesis Model for Jazz Improvisation

Open Access
|Feb 2022

Figures & Tables

tismir-5-1-87-g1.png
Figure 1

Overview of the generative model.

Table 1

Short overview of the Weimar Jazz Database.

The Weimar Jazz Database
Solos456
Performers78
Top PerformersColtrane (20), Davis (19), Parker (17), Rollins (13), Liebman (11), Brecker (10), Shorter (10), S. Coleman (10)
StylesTraditional (32), swing (66), bebop (56), cool (54), hardbop (76), postbop (147), free (5)
Instrumentsts (158), tp (101), as (80), tb (26), ss (23), other (68)
Time range1925–2009
Tone events200,809
Phrases11,802
Mid-level units15,402 (containing 5.2 WBA atoms on average)
WBA atoms80,123 (average length: 2.3 intervals)
Table 2

Overview of WBA atoms.

TypeSubtype/SymbolDescription
ScalesDiatonic (D)
Chromatic (C)
Diatonic scale
Chromatic scale
ApproachesFTwo intervals approaching a target pitch with a direction change (e. g., –2 +1)
TrillsTTwo alternating pitches
ArpeggiosSimple (A)
Jump (J)
Sequence of thirds
Sequence of intervals larger than a third
RepetitionsR
X atomsX
Link (L)
Miscellaneous category
X atoms of length 1
Table 3

Chord-scales used in the current model. Scale contents are given as pitch class vectors with 0 representing the root of the chord.

Chord TypeScalesScale content
maj, maj7Ionian[0, 2, 4, 5, 7, 9, 11]
min, min7Dorian[0, 2, 3, 5, 7, 9, 10]
7Mixolydian[0, 2, 4, 5, 7, 9, 10]
Major Blues[0, 2, 3, 4, 7, 9]
Mixolydian ♯11[0, 2, 4, 6, 7, 9, 10]
Altered Scale[0, 1, 3, 4, 6, 8, 10]
m7b5Locrian[0, 1, 3, 5, 6, 8, 10]
Phrygian[0, 1, 3, 5, 7, 8, 10]
o, o7Octatonic Scale[0, 2, 3, 5, 6, 8, 9, 11]
tismir-5-1-87-g6.png
tismir-5-1-87-g2.png
Figure 2

Example of a generated solo over an F-blues chord sequence, used in the evaluation (Algorithm-1-Original).

Table 4

Stimuli used for the evaluation. In column Performance Type specifics of the interpretation are given. Deadpan MIDI: Fully-quantized MIDI without dynamics; MIDI with microtiming: MIDI with semiautomatically added microtiming (swing); Audio: Recorded audio; Converted audio-to-MIDI: recorded audio converted to MIDI with Melodyne, keeping microtiming and dynamics; Recorded MIDI: human-played MIDI with microtiming and dynamics.

IdSolo IDGeneratorPerformance TypeSolo Sound
1Algorithm-1-OriginalWBA-MLU-AlgorithmDeadpan MIDItenor sax
2Algorithm-1-ImprovedWBA-MLU-Algorithm/AUT2MIDI with microtimingtenor sax
3Algorithm-2-OriginalWBA-MLU-AlgorithmDeadpan MIDItenor sax
4WJD-Sonny RollinsSonny Rollins (“Vierd Blues”)MIDI with microtimingtenor sax
5WJD-Miles DavisMiles Davis (“Vierd Blues”)MIDI with microtimingtenor sax
6WJD-Charlie ParkerCharlie Parker (“Billie’s Bounce”)MIDI with microtimingtenor sax
7Student-BeginnerBeginnerMIDI with microtimingtenor sax
8Student-IntermediateIntermediateMIDI with microtimingtenor sax
9Student-AdvancedAdvancedMIDI with microtimingtenor sax
10Student-GraduatedGraduatedMIDI with microtimingtenor sax
11Author-OriginalAUT2Audioe-guitar
12Author-MIDIAUT2Converted audio-to-MIDItenor sax
13Author-OriginalAUT1Audiopiano
14Author-MIDIAUT1Recorded MIDItenor sax
tismir-5-1-87-g3.png
Figure 3

Boxplot of MUSICALITY (A) and COMPLEXITY (B) values for all solos, separately for rater expertise. Left panels: Jazz experts; right panels: non-experts; blue: algorithmic solos; brown: author (MIDI) solos; yellow: student solos; green: author (original) solos; violet: WJD solos.

tismir-5-1-87-g4.png
Figure 4

Boxplot of liking values for sources of solos, separately for rater expertise. Left boxes: jazz experts, right: non-experts.

tismir-5-1-87-g5.png
Figure 5

Recognition accuracy of all 14 stimuli by expertise level. Left: all, middle: jazz experts, right: no jazz experts. Error bars are 95% confidence interval of proportion.

Table 5

Pearson’s correlation coefficients for MUSICALITY (MUS), COMPLEXITY (COMP), and artificial (Item 10). All correlations p ≤ .001.

MUSCOMPartificial
MUS1.000.44–0.59
COMP0.441.00–0.18
artificial–0.59–0.181.00
Table 6

Recognition accuracy for computer and human generated solos by experts and non-experts.

Solo GeneratorExpertiseAccuracy
AlgorithmExpert.644
Non-expert.417
HumanExpert.536
Non-expert.447
DOI: https://doi.org/10.5334/tismir.87 | Journal eISSN: 2514-3298
Language: English
Submitted on: Feb 26, 2021
Accepted on: Nov 24, 2021
Published on: Feb 3, 2022
Published by: Ubiquity Press
In partnership with: Paradigm Publishing Services
Publication frequency: 1 issue per year

© 2022 Klaus Frieler, Wolf-Georg Zaddach, published by Ubiquity Press
This work is licensed under the Creative Commons Attribution 4.0 License.