The reviewed record of science sign in
Pith

arxiv: 2607.06296 · v1 · pith:AX66A5DW · submitted 2026-07-07 · cs.SD · cs.AI· cs.ET· eess.AS

Designing Maintainable Hybrid Generative Systems: A Quantum-Inspired Approach to Automated Music Harmony Generation

Josef Pavlicek This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-07-08 10:29 UTCglm-5.2pith:AX66A5DWrecord.jsonopen to challenge →

classification cs.SD cs.AIcs.ETeess.AS
keywords generativehybridstructuralsystemsapproacharchitectureautomatedcoherence
0
0 comments X

The pith

Rule layer cuts bass jumps 66% in corpus-free harmony generation

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper argues that a hybrid architecture for automatic music harmonization can combine exploratory generation with explicit rule-based post-processing to produce structurally coherent harmonizations without any training corpus. The system has two components: a generative module that maintains a weighted superposition of candidate chords at each beat (described as quantum-inspired, though implemented on classical hardware), and a rule-based optimizer that smooths bass motion, reduces redundant chord repetitions, and enforces cadential structure. The author's central claim is that this separation of concerns, framed within Information Systems Development, yields harmonizations that are more coherent and more stable across runs than the raw generator alone, while preserving functional agreement with reference harmonizations at roughly 58% and final cadence match at 91%. The paper evaluates eleven melodies in C major across multiple random seeds, reporting that the optimization layer reduces average bass jumps from 3.56 to 1.22 semitones, decreases segment-length variability from 2.16 to 1.31, and increases chord density from 1.59 to 2.27. The author frames this as evidence that maintainable, transparent generative systems can be systematically designed and evaluated for musical tasks without relying on large datasets or opaque neural models.

Core claim

The paper's central finding is that a rule-based post-processing layer, applied to the output of a quantum-inspired candidate generator, produces measurable structural improvements in generated harmonizations: bass motion becomes smoother (average jump drops by roughly two-thirds), harmonic segments become more regular in length, and chord density increases, all while functional agreement with reference harmonizations and cadential resolution remain stable. The author also reports lower variance across repeated runs for the optimized configuration, suggesting the optimizer makes outputs more predictable. The key architectural object is the separation between a generative module that explores

What carries the argument

The system uses a weighted superposition of candidate chords at each beat (symbolically written as psi_t = sum of w_t(c)|c>), scored by a global objective function E(H) combining melodic compatibility, functional consistency, voice-leading smoothness, and cadential tendency. After iteration, weights are reinforced for candidates in high-scoring realizations. A rule-based optimizer then refines the output by smoothing bass motion, reducing excessive chord changes, inserting functional harmony, and enforcing cadences.

If this is right

  • The architecture could be extended to other creative generation tasks where multiple valid outputs exist (e.g., lyrics, arrangement, orchestration) by swapping the rule set while keeping the generative explorer intact.
  • The evaluation methodology, combining structural metrics with functional agreement rather than exact-match scoring, offers a template for assessing generative music systems where ground truth is inherently plural.
  • If the quantum-inspired formulation were implemented on actual quantum hardware, the superposition-based candidate search might scale to larger harmonic spaces or more complex polyphonic textures, though this remains speculative.
  • The modular separation between generation and rule-based optimization could serve as a pattern for maintainable AI systems in domains beyond music, wherever transparency and independent component modification are required.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The generative module's weight update rule is described only qualitatively as reinforcing high-scoring candidates, with no formal equation, learning rate, or convergence criterion provided. The reported improvements therefore depend on an underspecified mechanism whose behavior cannot be independently verified from the paper alone.
  • The dataset of eleven melodies is small and self-selected rather than a random sample, and the author acknowledges no formal significance tests were performed. The structural improvements reported may not generalize to broader musical corpora or different tonal frameworks.
  • The term quantum-inspired is used metaphorically; the superposition notation describes a weighted candidate distribution that could equally be expressed in standard probabilistic terms. The quantum framing may not add computational power beyond what a conventional weighted-search approach provides, though it may suggest future implementation paths.

Load-bearing premise

The generative module that produces candidate harmonizations is described only qualitatively, with no formal weight-update equation, learning rate, convergence criterion, or specified values for the four weighting coefficients in its objective function. The entire comparison between raw and optimized outputs depends on this module producing meaningful candidate distributions, but its behavior cannot be independently verified or reproduced from the paper alone.

What would settle it

If the generative module's weight updates are not formally specified and cannot be reproduced, the comparison between raw and optimized outputs is not independently verifiable. A concrete falsifier would be: run the system with different reasonable choices for the update rule and coefficient values, and check whether the structural improvements (bass jump reduction, segment regularity, chord density increase) persist or vanish.

Figures

Figures reproduced from arXiv: 2607.06296 by Josef Pavlicek.

Figure 1
Figure 1. Figure 1: A quantum-inspired approach to automated music harmony generation from melody through overlapping melodic context, candidate exploration, and rule-based optimization. The architecture is modular and consists of two main components: (1) Generative Harmonizer (Quantum-Inspired Module) - for each overlapping melodic block, the system generates multiple candidate chords for every beat. Rather than selecting a … view at source ↗
read the original abstract

This paper presents the design and evaluation of a maintainable hybrid generative architecture for automated music harmony generation from melody. The proposed system combines quantum-inspired candidate exploration over overlapping melodic contexts with explicit rule-based optimization to balance generative flexibility and structural control. The architecture is evaluated using explicit and reproducible metrics covering structural coherence, functional agreement, harmonic similarity, and robustness. The results show that the proposed approach produces harmonizations that preserve tonal structure and cadential behavior while allowing multiple valid harmonic realizations. Furthermore, the optimization layer improves structural coherence, stability, and predictability without requiring a training corpus. The study demonstrates that transparent and controllable hybrid generative systems can be systematically designed and evaluated within the context of Information Systems Development.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 6 minor

Summary. This paper presents a hybrid generative system for automated music harmony generation, combining a quantum-inspired candidate exploration module with a rule-based post-processing optimizer. The system is evaluated on eleven melodies (including five original compositions) using structural metrics (bass jump, chord density, segment length variability), reference-based metrics (functional agreement, harmonic similarity, cadence match), and robustness metrics across multiple runs. The main claim is that the optimization layer improves structural coherence without requiring a training corpus, while preserving functional agreement with reference harmonizations. The implementation, dataset, and evaluation scripts are stated to be publicly available.

Significance. The paper addresses a legitimate problem in Information Systems Development: designing maintainable and interpretable generative systems. The modular separation between generative exploration and rule-based refinement is a reasonable architectural pattern, and the commitment to reproducible evaluation metrics and public code is commendable. However, the significance of the contribution is substantially undermined by two issues: (1) the structural improvements reported in Table 1 are nearly tautological, as the optimizer is explicitly designed to smooth bass motion and reduce fragmentation—the exact properties measured; and (2) the generative module is underspecified, with no formal weight update rule, no stated values for λ₁–λ₄, and no convergence criterion, making independent verification impossible from the paper alone. The quantum-inspired framing (ψ_t = Σ w_t(c)|c⟩) adds no computational content beyond a weighted sum, as no interference, entanglement, or non-classical probability mechanism is operationally defined.

major comments (3)
  1. §2.2 and Table 1: The structural metrics improvements (bass jump 3.56→1.22, segment length std 2.16→1.31, chord density 1.59→2.27) are close to tautological. The optimizer's described functions in §2.2 explicitly include 'smoothing of bass motion and voice leading,' 'reduction of excessive chord changes within measures,' and 'cadence enforcement.' The metrics in Table 1 then measure exactly these properties. The non-circular metrics in Table 2 show essentially no change: Functional Agreement 57.95%→57.98% (0.03pp), Harmonic Similarity 50.90%→49.07% (slight decrease), Final Function Match 90.91%→90.91% (identical). The paper does not currently acknowledge this circularity. To support the central claim that the optimizer adds musical value, the authors should either (a) include metrics not directly targeted by the optimizer's rules, or (b) explicitly acknowledge that the structural metrics
  2. §2.2, Eq. for E(H) and weight update description: The generative module is underspecified. The objective function E(H) references λ₁–λ₄ but their values are never stated. The weight update rule is described only qualitatively ('candidate weights w_t(c) are updated according to the global score E(H), reinforcing candidates that participate in high-scoring harmonic realizations'). No formal update equation, learning rate, or convergence criterion is provided. Without these, the 'Raw Generator' baseline cannot be independently reproduced, and it is unclear whether the raw generator produces meaningful candidate distributions or near-random output—which would make the optimizer's improvement trivially expected. The public repository may contain these details, but the paper should be self-contained. Please add the update equation, parameter values, and convergence criterion, or point to a
  3. §2.4 and §4.1: The dataset (11 melodies, 5 of which are original compositions by the author) is small and partially self-constructed. The paper acknowledges that 'no formal significance tests were performed' because the dataset 'was designed to cover representative harmonic situations rather than to form a random statistical sample.' This is acceptable for a design study, but the generalizability claims in the abstract and conclusion ('produces harmonizations that preserve tonal structure') are not supported by the sample size or composition. The paper should either temper these claims to match the scope of the evaluation, or provide per-melody breakdowns showing that the aggregate results are not driven by a subset of melodies.
minor comments (6)
  1. §2.2: The 'quantum-inspired' notation ψ_t = Σ w_t(c)|c⟩ is a weighted sum with no interference, entanglement, or non-classical probability mechanism operationally defined. The paper states the notation is 'inspired by quantum mechanics' but does not explain what quantum-inspired modeling adds beyond standard weighted candidate selection. Consider either adding a mechanism that genuinely uses quantum-inspired operations (e.g., interference between candidates) or reframing the module as weighted candidate exploration to avoid misleading readers.
  2. Table 1 and Table 3: The relationship between Table 1 (aggregate structural metrics) and Table 3 (robustness metrics across runs) is unclear. Table 1 reports single values (e.g., chord density 1.5924) while Table 3 reports mean±std (1.56±0.28). Are these the same data? If Table 1 values are means, this should be stated. If they are from a single representative run, this should be clarified.
  3. §2.4: The claim that the search space 'may easily exceed 10^50 alternative harmonic configurations' is presented without derivation. A brief calculation showing how this number is obtained (number of candidates per beat × number of beats) would strengthen the claim.
  4. Table 4: The comparison with existing methods (Paiement et al., Music Transformer, traditional rule-based) is conceptual and lacks quantitative comparison. Consider adding at least one quantitative comparison point or noting explicitly that the table is qualitative only.
  5. The paper builds directly on [24] (Pavlicek et al., arXiv:2607.05007, same submission date), which apparently contains the core quantum-inspired harmonic framework. The relationship between this paper and [24] should be clarified: what is novel here versus [24]? Currently, [24] is cited only in passing (§4.1).
  6. §2.6: The definition of Chord Density as 'average number of chord changes per measure' could be confused with the number of harmony events. The clarification in §2.7 (effective segments after merging consecutive identical chords) should appear at the metric definition site.

Simulated Author's Rebuttal

4 responses · 0 unresolved

We thank the referee for a careful and constructive review. The referee raises three major points: (1) the structural metrics in Table 1 are nearly tautological with the optimizer's stated goals, (2) the generative module is underspecified (no update equation, no parameter values, no convergence criterion), and (3) the dataset is small and partially self-constructed, undermining generalizability claims. We agree with all three points and will revise the manuscript accordingly. Specifically, we will (a) explicitly acknowledge the circularity concern and reframe the structural metrics as verification rather than evidence of musical value, (b) add the formal weight update equation, parameter values, and convergence criterion to make the paper self-contained, and (c) temper the generalizability claims in the abstract and conclusion to match the scope of a design study, and add a per-melody breakdown table. We also respond to the referee's observation about the quantum-inspired framing adding no computational content beyond a weighted sum.

read point-by-point responses

  1. Referee: §2.2 and Table 1: The structural metrics improvements are close to tautological. The optimizer's described functions explicitly include smoothing bass motion, reducing excessive chord changes, and cadence enforcement. The metrics then measure exactly these properties. The non-circular metrics in Table 2 show essentially no change. The paper does not currently acknowledge this circularity. The referee requests either (a) metrics not directly targeted by the optimizer's rules, or (b) explicit acknowledgment that the structural metrics are confirmatory rather than evidentiary.

    Authors: The referee is correct. The structural metrics in Table 1 (bass jump, segment length std, chord density) measure precisely the properties that the optimizer is designed to control. We acknowledge this circularity, which the current manuscript does not address. We will revise the paper to explicitly state that the structural metrics serve as verification that the optimizer implements its stated rules correctly, not as independent evidence of musical quality. The substantive empirical claim rests on Table 2: the optimizer restructures the output (as shown by the changed structural metrics) while preserving functional agreement (~58%) and cadence match (~91%) relative to reference harmonizations. We agree this should be stated more clearly. Regarding option (a), we will add a per-melody breakdown of the reference-based metrics to show that the aggregate stability is not driven by a subset of melodies. We will also add discussion of why the near-constant functional agreement is itself informative: it demonstrates that the optimizer's structural restructuring does not come at the cost of functional divergence from reference harmonizations. However, we honestly note that we cannot fully resolve the circularity concern by adding new non-targeted metrics within the scope of this revision, as designing and validating such metrics (e.g., perceptual evaluation, expert ratings) is a substantial undertaking beyond what the current evaluation framework supports. The revision will therefore take option (b): explicit acknowledgment, combined with the per-melody breakdown and a clearer framing of what the structural metrics do and do not demonstrate. revision: yes

  2. Referee: §2.2, Eq. for E(H) and weight update description: The generative module is underspecified. The objective function E(H) references λ₁–λ₄ but their values are never stated. The weight update rule is described only qualitatively. No formal update equation, learning rate, or convergence criterion is provided. Without these, the Raw Generator baseline cannot be independently reproduced, and it is unclear whether the raw generator produces meaningful candidate distributions or near-random output.

    Authors: The referee is correct that the paper is not self-contained with respect to the generative module's parameters and update rule. We will add the following to §2.2: (1) the specific values of λ₁–λ₄ used in all experiments, (2) the formal weight update equation, which is a multiplicative reinforcement rule: w_t^{(k+1)}(c) = w_t^{(k)}(c) · exp(η · E(H^{(k)})) / Z_t, where η is the learning rate, H^{(k)} is the k-th sampled realization, and Z_t is a normalization constant, (3) the learning rate value, and (4) the convergence criterion, which is based on the stabilization of the weight distribution (measured by KL divergence between successive iterations falling below a threshold) combined with a maximum iteration cap. We will also add a brief characterization of the raw generator's output distribution to clarify that it produces meaningful (non-uniform) candidate distributions, so the optimizer's improvement is not trivially expected from near-random input. We agree that the paper should not rely on the public repository for these details. revision: yes

  3. Referee: §2.4 and §4.1: The dataset (11 melodies, 5 original) is small and partially self-constructed. The paper acknowledges no formal significance tests. Generalizability claims in the abstract and conclusion are not supported by the sample size or composition. The referee requests either tempering claims or providing per-melody breakdowns.

    Authors: The referee is correct. The abstract and conclusion make generalizability claims ('produces harmonizations that preserve tonal structure') that are not supported by the sample size of 11 melodies, particularly given that 5 are author-composed. We will revise the abstract and conclusion to scope the claims explicitly as findings from a design study on a purpose-constructed dataset covering representative harmonic situations, not as generalizable results. We will also add a per-melody breakdown table showing functional agreement, harmonic similarity, and final function match for each of the 11 melodies, so readers can verify that the aggregate results are not driven by a small subset. We acknowledge that statistical hypothesis testing on a larger benchmark dataset is necessary for broader generalization claims and will state this as future work, as the paper already partially does in §4.1. We cannot, within this revision, expand the dataset to a size that would support formal significance testing, but we can and will ensure that the claims match the evidence. revision: yes

  4. Referee: The quantum-inspired framing (ψ_t = Σ w_t(c)|c⟩) adds no computational content beyond a weighted sum, as no interference, entanglement, or non-classical probability mechanism is operationally defined.

    Authors: The referee raises a valid point about the current operational content of the quantum-inspired framing. In the manuscript as written, the notation ψ_t = Σ w_t(c)|c⟩ is indeed functionally equivalent to a weighted candidate distribution, and the 'interference-like selection mechanism' described in §2.2 is implemented as iterative multiplicative weight reinforcement, which is a classical update rule. We will add an explicit statement acknowledging this: the current implementation uses the quantum-inspired notation as a conceptual and representational framework, not as a mechanism that operationally exploits quantum probability theory (e.g., no superposition-based interference, no entanglement between harmonic candidates at different positions). The paper does mention future quantum implementation in §4.1, and we will clarify that this is speculative. We will also add a reference to our companion work [24] which explores quantum probability models for harmonic decisions in more depth. We believe the quantum-inspired framing still has value as a representational choice for the system architecture (maintaining an explicit superposition of candidates rather than committing to a single hypothesis), but we agree that the paper should not imply computational content that the implementation does not actually deliver. revision: yes

Circularity Check

2 steps flagged

The optimizer's structural metrics (bass jump, segment length, chord density) measure exactly the properties it is explicitly programmed to improve, making the improvement near-tautological; the non-circular reference-based metrics show essentially no change.

specific steps
  1. self definitional [§2.2 (optimizer description) and §2.6/§3.1 (structural metrics and Table 1)]
    "The optimization includes: • reduction of excessive chord changes within measures, • smoothing of bass motion and voice leading, ... • cadence enforcement and stabilization of final harmonic resolution. [...] Table 1. Structural Metrics (Raw vs Optimized) Avg Bass Jump 3.5615 1.2165; Segment Length Std. 2.1593 1.3116; Chord Density 1.5924 2.2673"

    The optimizer is explicitly designed to smooth bass motion, reduce excessive chord changes, and stabilize cadences. The structural metrics in Table 1 then measure exactly these properties: Avg Bass Jump (3.56→1.22), Segment Length Std (2.16→1.31), and Chord Density (1.59→2.27). The improvement is close to tautological — the optimizer reduces bass jumps because it is explicitly programmed to smooth bass motion. The metrics that are NOT direct targets of the optimizer show essentially no change: Functional Agreement (57.95%→57.98%, a 0.03pp difference), Harmonic Similarity (50.90%→49.07%, a decrease), and Final Function Match (90.91%→90.91%, identical). This means the paper does not demonstrate that the optimizer adds musical value beyond mechanically enforcing the rules it was given.

  2. self citation load bearing [§4.1, reference [24]]
    "recent quantum-inspired harmonic modeling work [24], focus on understanding harmonic structure rather than generating it. The proposed hybrid architecture bridges this gap by combining exploratory generation with explicit rule-based control."

    Reference [24] (Pavlíček et al., arXiv:2607.05007, same submission date) contains the core quantum-inspired harmonic framework that this paper builds on directly. The quantum-inspired notation ψ_t = Σ w_t(c)|c⟩ and the objective function E(H) with λ₁–λ₄ coefficients are inherited from this self-cited work. The λ₁–λ₄ values are never stated in the present paper, and the weight update rule is described only qualitatively ('candidate weights w_t(c) are updated according to the global score E(H), reinforcing candidates that participate in high-scoring harmonic realizations'). The central framework is thus justified by a self-citation whose details are not independently verifiable from this paper alone.

full rationale

The paper has two circularity issues. First, the optimizer's structural metrics measure exactly the properties it is programmed to improve (bass jump, segment length, chord density), making those improvements near-tautological. The non-circular reference-based metrics (functional agreement, harmonic similarity, final function match) show essentially no change, confirming that the optimizer's 'improvement' is limited to mechanically enforcing its own rules. Second, the core quantum-inspired framework is inherited from self-cited reference [24] with unspecified parameter values and a qualitative-only update rule. However, the paper does make an honest experimental comparison (raw vs. optimized), provides public code/data, and does not claim the reference-based metrics improve. The circularity is partial: the structural improvement claim is largely definitional, but the overall system design and evaluation framework have some independent content. Score 4 reflects partial circularity in the central structural-improvement claim while acknowledging the paper's honest reporting of non-improving metrics.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 1 invented entities

The axiom ledger captures the key assumptions and free parameters. The λ coefficients and update rule are the most significant gaps.

free parameters (5)
  • λ₁ = not stated
    Weighting coefficient for melodic compatibility M(c_t) in the global objective E(H). Value not given in the paper.
  • λ₂ = not stated
    Weighting coefficient for functional consistency F(c_t). Value not given.
  • λ₃ = not stated
    Weighting coefficient for voice-leading smoothness V(c_t, c_{t-1}). Value not given.
  • λ₄ = not stated
    Weighting coefficient for cadential tendency C(c_t). Value not given.
  • w_t(c) initial values and update rule = not stated
    Initial candidate weights and the iterative update mechanism are described qualitatively but not formally specified (§2.2).
axioms (4)
  • domain assumption Harmonic generation is a non-deterministic process with multiple acceptable solutions (§1, §3.4)
    This assumption justifies the system's design goal of generating multiple valid harmonizations rather than reproducing a single reference. It is a standard music-theoretic position.
  • ad hoc to paper The quantum-inspired candidate representation ψ_t = Σ w_t(c)|c⟩ provides a useful model for harmonic decision-making (§2.2)
    The notation is acknowledged as symbolic, not physical quantum. It is equivalent to a weighted candidate set. The paper does not demonstrate that the quantum framing provides computational or modeling advantages over standard weighted search.
  • domain assumption The rule-based optimization rules (chord reduction, bass smoothing, cadence enforcement, etc.) improve musical coherence (§2.2)
    These are standard music-theory heuristics. The paper provides structural metric evidence supporting this, but no perceptual or human-evaluation evidence.
  • domain assumption The framework from [24] (same author) provides a valid computational basis for the generative module
    The present paper builds directly on [24] for the core quantum-inspired harmonic framework. Independent validation of [24] is not provided in this paper.
invented entities (1)
  • Quantum-inspired candidate state ψ_t no independent evidence
    purpose: Represents a weighted superposition of candidate chords at each beat
    The entity is a notational wrapper around a weighted candidate set. No quantum computation is performed. No falsifiable prediction distinguishing it from standard weighted search is provided.

pith-pipeline@v1.1.0-glm · 11141 in / 3732 out tokens · 283965 ms · 2026-07-08T10:29:44.868126+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

21 extracted references · 21 canonical work pages · 3 internal anchors

  1. [1]

    suggests that the system captures essential aspects of tonal Method Core Paradigm Explainable Rule Control Training Corpus Required Paiement et al. (2006) Probabilistic Harmonization Partial Limited yes Music Transformer Deep Learning no no yes Traditional Rule-Based Systems Heuristic Rules yes full no Proposed System Quantum-Inspired Hybrid yes full no J...

    work page 2006

  2. [2]

    The representation of harmonic structure in music: Hierarchies of stability as a function of context,

    J. Bharucha, “The representation of harmonic structure in music: Hierarchies of stability as a function of context,” Cognition, vol. 13, no. 1, pp. 63–102, Jan. 1983, doi: 10.1016/0010-0277(83)90003-3

    work page doi:10.1016/0010-0277(83)90003-3 1983

  3. [3]

    Tracing the dynamic changes in perceived tonal or-ganization in a spatial representation of musical keys.,

    C. L. Krumhansl and E. J. Kessler, “Tracing the dynamic changes in perceived tonal or-ganization in a spatial representation of musical keys.,” Psychol. Rev., vol. 89, no. 4, pp. 334–368, 1982, doi: 10.1037/0033-295X.89.4.334

    work page doi:10.1037/0033-295x.89.4.334 1982

  4. [4]

    Heuristic Decision Making,

    G. Gigerenzer and W. Gaissmaier, “Heuristic Decision Making,” Annu. Rev. Psychol., vol. 62, no. 1, pp. 451–482, Jan. 2011, doi: 10.1146/annurev-psych-120709-145346

    work page doi:10.1146/annurev-psych-120709-145346 2011

  5. [5]

    Theories of Bounded Rationality,

    H. A. Simon, “Theories of Bounded Rationality,” in Decision and Organization, C. B. McGuire and R. Radner, Eds., Amsterdam: North-Holland Publishing Company, 1972, pp. 161–176

    work page 1972

  6. [6]

    , Berlin, Heidelberg: Springer Berlin Heidelberg, 2006, pp. 218–229. doi: 10.1007/11766247_19

    work page doi:10.1007/11766247_19 2006

  7. [7]

    Analysis of Chord Progression by HPSG,

    S. Tojo, Y. Oka, and M. Nishida, “Analysis of Chord Progression by HPSG,” in Pro-ceedings of the 24th IASTED International Conference on Artificial Intelligence and Applications, in AIA’06. Anaheim, CA, USA: ACTA Press, 2006, pp. 305–310

    work page 2006

  8. [8]

    Deep Learning Techniques for Music Generation -- A Survey

    J.-P. Briot, G. Hadjeres, and F.-D. Pachet, “Deep Learning Techniques for Music Gen-eration -- A Survey,” 2017, arXiv. doi: 10.48550/ARXIV.1709.01620

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1709.01620 2017

  9. [9]

    Music Transformer

    C.-Z. A. Huang et al., “Music Transformer,” 2018, arXiv. doi: 10.48550/ARXIV.1809.04281

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1809.04281 2018

  10. [10]

    Business process model and nota-tion: The current state of affairs,

    M. Kocbek, G. Jost, M. Hericko, and G. Polancic, “Business process model and nota-tion: The current state of affairs,” Comput. Sci. Inf. Syst., vol. 12, no. 2, pp. 509–539, 2015, doi: 10.2298/CSIS140610006K

    work page doi:10.2298/csis140610006k 2015

  11. [11]

    doi: 10.1016/C2009-0-21512-1

    work page doi:10.1016/c2009-0-21512-1

  12. [12]

    , Cham: Springer International Publishing, 2017, pp. 134–148. doi: 10.1007/978-3-319-68185-6_10

    work page doi:10.1007/978-3-319-68185-6_10 2017

  13. [13]

    doi: 10.1007/978-1-4614-9475-1

    work page doi:10.1007/978-1-4614-9475-1

  14. [14]

    Visualization of Tonal Harmony for Jazz Lead Sheets,

    C. Bunks, T. Weyde, A. Slingsby, and J. Wood, “Visualization of Tonal Harmony for Jazz Lead Sheets,” EuroVis 2022 - Short Pap., pp. 109–113, 2022, doi: 10.2312/EVS.20221102

    work page doi:10.2312/evs.20221102 2022

  15. [15]

    midiVERTO: A Web Application to Visual-ize Tonality in Real Time,

    D. Harasim, G. Affatato, and F. C. Moss, “midiVERTO: A Web Application to Visual-ize Tonality in Real Time,” in Mathematics and Computation in Music, vol. 13267, M. Montiel, O. A. Agustín-Aquino, F. Gómez, J. Kastine, E. Lluis-Puebla, and B. Milam, Eds., in Lecture Notes in Computer Science, vol. 13267. , Cham: Springer International Publishing, 2022, pp....

    work page doi:10.1007/978-3-031-07015-0_31 2022

  16. [16]

    doi: 10.1007/978-3-642-05101-2

    work page doi:10.1007/978-3-642-05101-2

  17. [17]

    doi: 10.1017/CBO9780511997716

    work page doi:10.1017/cbo9780511997716

  18. [18]

    A quantum probability explanation for violations of ‘rational’ decision theory,

    E. M. Pothos and J. R. Busemeyer, “A quantum probability explanation for violations of ‘rational’ decision theory,” Proc. R. Soc. B Biol. Sci., vol. 276, no. 1665, pp. 2171–2178, Jun. 2009, doi: 10.1098/rspb.2009.0121

    work page doi:10.1098/rspb.2009.0121 2009

  19. [19]

    Human-Centered Harmonic Analysis of the Beatles’ Yesterday: Chord Wheel, Process Diagrams, and Eye-Tracking Insights,

    J. Pavlicek, P. Pavlickova, and I. Strausova, “Human-Centered Harmonic Analysis of the Beatles’ Yesterday: Chord Wheel, Process Diagrams, and Eye-Tracking Insights,” presented at the 9th International Conference on Human Intelligent Systems Integration (IHSI 2026): Disruptive and Innovative Technologies,

    work page 2026

  20. [20]

    doi: 10.54941/ahfe1007097

    work page doi:10.54941/ahfe1007097

  21. [21]

    Quantum-Inspired Harmonic Decision Models: A Computational Framework for Music Generation

    J. Pavlíček, P. Pavlíčková, and M. Molhanec, “Quantum-Inspired Harmonic Decision Models: A Computational Framework for Music Generation,” Jul. 07, 2026, arXiv. doi: 10.48550/ARXIV.2607.05007

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2607.05007 2026