Pith. sign in

REVIEW 3 major objections 5 minor 42 references

Calliphony: A Calligraphy-Driven Interface for Real-Time Generative Music Performance

T0 review · 3 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

Pith's one-line read Calliphony establishes that brush motion, measured only as rotational speed, can serve as an expressive external control layer for a real-time symbolic music generation model.

desk verdict A credible NIME systems demo with a genuinely new external-trigger control of Notochord, but the key integration with the model's timing state is under-specified and unmeasured. read the letter →

arxiv 2608.03040 v1 pith:OFSLE2Z7 submitted 2026-08-04 cs.SD

classification cs.SD
keywords calligraphyreal-timegenerativemusicsymbolicgenerationgesturalcontrolNotochordMIDIperformancehuman-AIco-creationmultimodal
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Calliphony is a live performance system that lets a calligrapher drive a real-time generative music model with brush motion. Its central claim is that a continuous speed signal from a gyroscope mounted on the brush, integrated over time, can decide when the model emits each next note, so faster writing produces denser melody and stillness produces silence. The same speed value, with threshold-based activation, brings chord, bass, and sub-melody layers in and out, creating a full arrangement from gesture alone. If this holds, calligraphy becomes an audiovisual, AI-assisted co-performance medium rather than only a visual art, and non-musical gestures can serve as an external control layer for symbolic music generation.

What carries the argument

The central mechanism is the distance-integral trigger: a continuous speed value derived from the brush's tri-axial gyroscope rotation is integrated over time in Max, and each time the accumulated movement distance crosses a user-set threshold, the system queries Notochord to predict and output the next melody note. The same speed signal, through threshold-based onset and stop values with hysteresis, switches the chord, bass, and sub-melody channels on and off. This mechanism is what converts calligraphic motion into note-onset density, harmonic layering, and textural change.

What would settle it

Record the same gyroscope speed profile with a brush and with an ordinary stick, and also vary brush pressure while holding rotation constant; if both pairs of trials produce identical music, then rotation speed alone carries the expressivity, and the calligraphy-specific sensor dimensions contribute nothing to the control signal.

Watch

Extended reading notes

Core claim

On its own terms, the paper's discovery is that brush motion does not need to enter the neural network as an input feature to control it; the motion can instead control when and under what constraints the model is queried. Calliphony converts tri-axial gyroscope rotation into a scalar speed value, integrates that value over time, and triggers Notochord to predict the next melody note whenever the accumulated movement distance reaches a preset threshold. Pitch and velocity remain the model's choices, but inter-onset interval is replaced by the performer's motion, making note density a direct function of writing speed. A second speed threshold with hysteresis switches chords, bass, and a sub-melody on or off, so acceleration thickens the texture and deceleration strips it back to a monophonic line or silence.

Load-bearing premise

The system assumes that a scalar speed value from the brush's rotation, integrated over time, is expressive enough to shape musical density and texture, even though it deliberately ignores pressure, contact area, ink flow, and brush-tip deformation.

Editorial extensions

If this is right

  • A calligrapher can perform a multi-track composition (melody, chords, bass, and sub-melody) using only brush speed, with no keyboard or sequencer input.
  • Pauses and slow passages become rests and sparse texture, so the phrasing of handwritten characters is translated directly into musical phrasing.
  • Because the control signal is a scalar accumulation of rotation speed, the same interaction layer can be attached to any moving object or body part without retraining the generative model.
  • Scale filtering and pitch-repeat suppression keep the generated melody tonally coherent even though the model was not trained for calligraphy-specific control.
  • The generative model can be swapped for another real-time MIDI model without redesigning the gesture interface, since the interface operates on query timing rather than model internals.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The authors do not test whether pressure, contact area, or brush-tip deformation would add musical expressivity if sensed; their own discussion identifies these as missing dimensions, so the system's ceiling is set by the speed-only sensor.
  • A natural extension is to invert the mapping and use accumulated distance to control pitch contour, scale choice, or timbre rather than onset density, a variation the authors do not explore.
  • The same distance-integral control could be applied to other time-based embodied practices, such as conducting, dance, or tai chi, where velocity and pauses carry phrase structure; this follows from the paper's architecture but is not claimed by it.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper presents Calliphony, a calligraphy-driven interface for real-time generative music performance. An M5Stack gyroscope attached to a calligraphy brush streams angular velocity to a Max patch, which derives a scalar speed value. This speed controls two mechanisms in a Python-based generative layer built around the Notochord autoregressive MIDI model: a distance-integral trigger that queries the model for the next melody note when accumulated motion reaches a threshold, and threshold-based activation of chord, bass, and sub-melody tracks. Output is routed to Ableton Live through LoopMIDI. The authors report an informal five-minute live performance at an art event and acknowledge that no formal recording or user study was conducted. The paper's contributions are framed as a performance-oriented prototype and a cross-modal performance scenario.

Significance. If the described integration works as stated, the system is a plausible and potentially useful NIME-style contribution: it uses an external, non-musical gestural signal to control the query timing and structural activation of an existing real-time generative model, rather than treating the model as an autonomous composer. The paper is concrete about the hardware/software pipeline, uses a publicly available model, and is honest about the absence of formal evaluation. The strength is the clear architectural description: Max for sensor processing, Python for model control, and Ableton for rendering, with OSC between components. The main weakness is that the central mechanism—replacing Notochord's internal inter-onset interval with an externally triggered query—is under-specified at the level of the model's hidden state and event feedback, which is load-bearing for the density-control claim. In addition, the 'low-latency' claim is unmeasured. With a precise description of the inference loop and at least a small quantitative or logged demonstration, the paper would substantiate its claims.

major comments (3)
  1. [§3.2.1] The core mechanism is under-specified: the paper states that 'the inter-onset interval is no longer decided by the model' and that the system triggers Notochord queries when a distance integral reaches a threshold, but it does not explain how Notochord's event representation and GRU hidden state are updated under this scheme. In Notochord, each event is a tuple of instrument, pitch, inter-onset interval, and velocity, and the model samples the IOI as part of the event; in the original auto mode, note endings are represented by zero-velocity events. If Calliphony discards the sampled IOI and query times are set externally, the hidden state must be updated with the actual elapsed time between triggers and with a representation of note-off/velocity-zero, but the paper does not specify which values are fed back. If the sampled IOI is fed back while the actual timing is externally controlled, the model's internal clock diverges from the performance timeline, and future pitch/chord samples come from an out-of-distribution hidden state; if the actual IOI is fed back, that should be stated explicitly. This is load-bearing for the central claim that brush motion controls note-onset density, and it must be resolved in a revised version.
  2. [Abstract and §3.3] The abstract calls the pipeline 'low-latency,' but no latency measurement is reported anywhere in the paper. For an interactive music performance system, the real-time claim should be supported by at least a round-trip latency measurement from sensor input to MIDI/audio output, or by CPU load and buffer-size data. Without this, the suitability of the system for live performance is asserted rather than demonstrated. The authors should either add such a measurement or explicitly qualify the claim as unquantified.
  3. [§4] The only reported evaluation is informal audience feedback from a single five-minute performance that was not recorded, and the paper explicitly acknowledges that this does not constitute user-study data. This is acceptable for a demo-oriented proceedings paper, but it means the functional claims—distance-integral density control, threshold-based layer activation, scale-based pitch constraints—are not verified. A short logged demonstration, such as a recorded sensor trace with the resulting MIDI output, or a simple offline analysis of event-density versus brush-speed correlation, would substantially strengthen the central claims.
minor comments (5)
  1. [§3.1] The method for obtaining a scalar speed value from accumulated rotation angles does not discuss gyroscope drift, offset calibration, or integration error; given that the signal is accumulated over time, a brief note on how drift was handled in practice would improve reproducibility.
  2. [§3.1 and §2.1] The paper emphasizes brush direction and turns as expressive features in calligraphy, but the control signal is a scalar speed value that discards directional information; please clarify whether direction is intentionally unused or whether future work plans to exploit it.
  3. [§1] There are several typographical spacing errors, including 'This paper presentsCalliphony' and 'poemShui Diao Ge Touon a water-writing cloth'; these should be corrected in a final pass.
  4. [§2.2] The sentence 'Early work such as Hé: Calligraphy as a Musical Interface extract computable features' has a subject-verb agreement error ('extract' should be 'extracts'); this is a minor presentational fix.
  5. [§4 and §7] The paper promises to release code and a video demo upon publication; if possible, the authors should include a link or a supplementary demo file in the revised version, since the unrecorded performance cannot otherwise be inspected.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the control mapping is by construction and Notochord is an external model; self-citations are background only.

full rationale

The paper is an engineering and prototype description rather than a derivation or empirical prediction. The claimed control mechanism in Section 3.2.1 is implemented by definition: integrating a speed signal and triggering a Notochord query at threshold crossings directly defines note-onset density; the paper does not present this relationship as a discovered prediction, so there is no fitted input renamed as a prediction. Notochord is an external, publicly released model [30], and the system uses its query and feed interfaces without fitting any parameters to the authors' own data. Thresholds and exclusion probabilities are user-adjustable interface parameters, not fitted constants. Self-citations such as [18,19,21,37,40] appear only as background on calligraphy generation and symbolic music modeling; none is load-bearing for the Calliphony pipeline, and no uniqueness theorem from the authors' prior work is invoked. The only substantive concern raised by the system description is whether overriding Notochord's inter-onset interval modality without specifying how the GRU hidden state accounts for actual elapsed time puts the model out of distribution; that is a correctness or under-specification risk, not circularity, because it does not make any output equal to an input by construction. Therefore no circular steps are identified.

Assumptions & free parameters 3 free parameters · 2 assumptions · 0 invented entities

The system relies on Notochord's documented behavior and on the assumption that rotational speed is an adequate proxy for writing energy. No novel entities are introduced; the off-the-shelf hardware and external model carry the assumptions.

free parameters (3)
  • distance-integral trigger threshold = unspecified, user-adjustable
    Preset threshold for accumulated rotational angle that fires each melody note; controls mapping from motion speed to note density (Section 3.2.1).
  • onset and stop thresholds for accompaniment = unspecified, user-adjustable
    Speed thresholds with hysteresis band activating chord, bass, and sub-melody layers (Section 3.2.2).
  • pitch-repeat exclusion probability = user-adjustable 0..1
    Probability of excluding the most recent pitch from the candidate set to avoid monotony (Section 3.2.1).
assumptions (2)
  • domain assumption Notochord's autoregressive GRU model, with the publicly released checkpoint, provides coherent symbolic generation when queried under constraints.
    The system's melodic and harmonic output quality depends on the pretrained Notochord model behaving as documented (Section 3.2).
  • domain assumption Integration of tri-axial angular velocity over time yields a scalar that is monotonically related to perceived writing energy.
    The trigger mapping assumes accumulated gyro angle is a faithful proxy for calligraphic motion intensity (Section 3.2.1).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Calliphony: A Calligraphy-Driven Interface for Real-Time Generative Music Performance." pith.science (2026). https://pith.science/paper/OFSLE2Z7

@misc{pith2026260803040,
  author       = {Pith},
  title        = {Pith review of: Calliphony: A Calligraphy-Driven Interface for Real-Time Generative Music Performance},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OFSLE2Z7}},
  note         = {Machine review of arXiv:2608.03040}
}
read the original abstract

While music generative models have recently gained significant attention, how they can be effectively integrated into live music performances still requires further exploration. This paper presents Calliphony, a calligraphy-driven interface for real-time generative music performance. Specifically, we build a low-latency pipeline that captures brush motion with an attachable sensor and maps it to control signals for real-time symbolic music generation. Using a generative model, the system produces multi-track MIDI in performance settings, while brush-derived control signals constrain event timing and activate additional musical layers. The generated melody is then extended with real-time harmony and additional voices, and finally rendered through a DAW for live staging. Calliphony contributes: (1) a performance-oriented prototype that uses calligraphic motion as an external control layer for a real-time symbolic music generation model, controlling note density, pitch constraints, and accompaniment-layer activation; and (2) a cross-modal performance scenario that extends calligraphy beyond a primarily visual practice into an audiovisual, AI-assisted setting.

Figures

Figures reproduced from arXiv: 2608.03040 by the authors.

Figure 1
Figure 1. Pipeline of the whole system. 3.1 Data Input We use the built-in gyroscope of an M5Stack2 (a modular devel￾opment board that integrates a microcontroller, a screen, and multiple sensors/expansion interfaces) to capture the brush’s rotational motion in space. We chose this brush-mounted gy￾roscope setup because it preserves the existing writing surface, requires no special paper, and allows the performer to maintain … view at source ↗
Figure 2
Figure 2. Brush with M5Stack [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Control UI for real-time parameter tuning and [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Live performance of Calliphony. interaction between the music generation system and the per￾former’s movements. Others were more concerned with the data acquisition during writing—for example, whether different stroke shapes and brush pressure influenced the generated …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 33 canonical work pages

  1. [1]

    Denk, Zalán Borsos, Jesse Engel, Mauro Verzetti, Antoine Caillon, Qingqing Huang, Aren Jansen, Adam Roberts, Marco Tagliasacchi, et al

    Andrea Agostinelli, Timo I. Denk, Zalán Borsos, Jesse Engel, Mauro Verzetti, Antoine Caillon, Qingqing Huang, Aren Jansen, Adam Roberts, Marco Tagliasacchi, et al. MusicLM: Generating music from text.arXiv preprint arXiv:2301.11325, 2023

  2. [2]

    BachDuet: A human-machine duet improvisation system

    Christodoulos Benetatos and Zhiyao Duan. BachDuet: A human-machine duet improvisation system. InLate-Breaking/Demo at the 20th International Society for Music Information Retrieval Conference (ISMIR), Delft, The Netherlands,

  3. [3]

    Phantom of utopia

    KAHEI CHENG, Irina Kruchinina, and Matin Esmaeili. Phantom of utopia. InProceedings of the International Conference on New Interfaces for Musical Expression, pages 142–145, 2024

  4. [4]

    Die schönheit der vergänglichkeit

    Se-Lien Chuang and Andreas Weixler. Die schönheit der vergänglichkeit. InProceedings of the International Conference on New Interfaces for Musical Expression, pages 70–73, 2024

  5. [5]

    MusicGen: Simple and controllable music generation.arXiv preprint arXiv:2306.05284, 2023

    Jade Copet, Felix Kreuk, Gabriel Synnaeve, Yossi Adi, et al. MusicGen: Simple and controllable music generation.arXiv preprint arXiv:2306.05284, 2023

  6. [6]

    Chinese calligraphy, 2008

    Dawn Delbanco. Chinese calligraphy, 2008. The Metropolitan Museum of Art, accessed: 2026-02-03

  7. [7]

    Rahul Dey and Fathi M. Salem. Gate-variants of gated recurrent unit (GRU) neural networks. In2017 IEEE 60th International Midwest Symposium on Circuits and Systems (MWSCAS), pages 1597–1600. IEEE, 2017

  8. [8]

    Jukebox: A generative model for music.arXiv preprint arXiv:2005.00341, 2020

    Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Rad- ford, and Ilya Sutskever. Jukebox: A generative model for music.arXiv preprint arXiv:2005.00341, 2020

Show all 42 references
  1. [9]

    MMM: Exploring conditional multi-track music generation with the transformer, 2020

    Jeff Ens and Philippe Pasquier. MMM: Exploring conditional multi-track music generation with the transformer, 2020

  2. [10]

    The four treasures of the study: Ink, inkstone, brush, and paper, 2019

    Howard Hotson. The four treasures of the study: Ink, inkstone, brush, and paper, 2019. University of Oxford, accessed: 2026-02-03

  3. [11]

    Music transformer.arXiv preprint arXiv:1809.04281, 2018

    Cheng-Zhi Anna Huang, Ashish Vaswani, Jakob Uszkoreit, Noam Shazeer, Ian Simon, Curtis Hawthorne, Andrew Dai, Matthew Hoffman, Monica Dinculescu, and Douglas Eck. Music transformer.arXiv preprint arXiv:1809.04281, 2018

  4. [12]

    Pop music transformer: Beat-based modeling and generation of expressive pop piano compositions

    Yu-Siang Huang and Yi-Hsuan Yang. Pop music transformer: Beat-based modeling and generation of expressive pop piano compositions. InProceedings of the 28th ACM International Conference on Multimedia, MM ’20, pages 1180– 1188, New York, NY, USA, 2020. Association for Computing ...

  5. [13]

    Self-cultivation through art: Chinese calligraphy and the body

    Ruyu Hung. Self-cultivation through art: Chinese calligraphy and the body. Educational Philosophy and Theory, 2021. Published online: 14 Sep 2021

  6. [14]

    Infilling piano performances

    Daphne Ippolito, Cheng-Zhi Anna Huang, Curtis Hawthorne, and Douglas Eck. Infilling piano performances. InNeurIPS Workshop on Machine Learning for Creativity and Design, 2018. Workshop paper / online supplement

  7. [15]

    Hé: Calligraphy as a musical interface

    Laewoo Kang and Hsin-Yi Chien. Hé: Calligraphy as a musical interface. InProceedings of the International Conference on New Interfaces for Musical Expression, pages 352–355, 2010

  8. [16]

    Collaboration and recursion: Reflections on calligraphy and feedback

    Iurii Kuzmin, Omar Al Kanawati, and Raul Masu. Collaboration and recursion: Reflections on calligraphy and feedback. InProceedings of the International Conference on New Interfaces for Musical Expression, pages 175–183, 2025

  9. [17]

    George E. Lewis. Too many notes: Computers, complexity and culture in Voyager.Leonardo Music Journal, 10:33–39, 2000

  10. [18]

    CalliffusionV2: Personalized natural calligraphy generation with flexible multi-modal control, 2024

    Qisheng Liao, Liang Li, Yulang Fei, and Gus Xia. CalliffusionV2: Personalized natural calligraphy generation with flexible multi-modal control, 2024

  11. [19]

    Calliffusion: Chinese calligraphy generation and style transfer with diffusion modeling

    Qisheng Liao, Gus Xia, and Zhinuo Wang. Calliffusion: Chinese calligraphy generation and style transfer with diffusion modeling. InProceedings of the International Conference on Computational Creativity, 2023

  12. [20]

    Moyun: A diffusion-based model for style-specific chinese calligraphy generation, 2024

    Kaiyuan Liu, Jiahao Mei, Hengyu Zhang, Yihuai Zhang, Xingjiao Wu, Daoguo Dong, and Liang He. Moyun: A diffusion-based model for style-specific chinese calligraphy generation, 2024

  13. [21]

    Polyffusion: A diffusion model for polyphonic score generation with internal and external controls

    Lejun Min, Junyan Jiang, Gus Xia, and Jingwei Zhao. Polyffusion: A diffusion model for polyphonic score generation with internal and external controls. arXiv preprint arXiv:2307.10304, 2023

  14. [22]

    Symbolic music generation with diffusion models

    Gautam Mittal, Jesse Engel, Curtis Hawthorne, and Ian Simon. Symbolic music generation with diffusion models. InProceedings of the International Society for Music Information Retrieval Conference (ISMIR), 2021

  15. [23]

    This time with feeling: Learning expressive musical performance.Neural Computing and Applications, 32:955–967, 2020

    Sageev Oore, Ian Simon, Sander Dieleman, Douglas Eck, and Karen Simonyan. This time with feeling: Learning expressive musical performance.Neural Computing and Applications, 32:955–967, 2020

  16. [24]

    Calligraphy

    Oxford Bibliographies. Calligraphy. Oxford Bibliographies: Chinese Studies. Retrieved February 8, 2026

  17. [25]

    The Continuator: Musical interaction with style.Journal of New Music Research, 32(3):333–341, 2003

    François Pachet. The Continuator: Musical interaction with style.Journal of New Music Research, 32(3):333–341, 2003

  18. [26]

    Colin Raffel and Daniel P. W. Ellis. Extracting ground-truth information from MIDI files: A MIDIfesto. InProceedings of the International Society for Music Information Retrieval Conference (ISMIR), pages 796–802, 2016

  19. [27]

    PopMAG: Pop music accompaniment generation

    Yi Ren, Jinzheng He, Xu Tan, Tao Qin, Zhou Zhao, and Tie-Yan Liu. PopMAG: Pop music accompaniment generation. InProceedings of the 28th ACM Inter- national Conference on Multimedia, MM ’20, pages 1190–1198. Association for Computing Machinery, 2020

  20. [28]

    A hierarchical latent vector model for learning long-term structure in music

    Adam Roberts, Jesse Engel, Colin Raffel, Curtis Hawthorne, and Douglas Eck. A hierarchical latent vector model for learning long-term structure in music. InProceedings of the 35th International Conference on Machine Learning (ICML), 2018

  21. [29]

    MIT Press, 1993

    Robert Rowe.Interactive Music Systems: Machine Listening and Composing. MIT Press, 1993

  22. [30]

    Notochord: A flexible probabilistic model for real-time MIDI performance.arXiv preprint arXiv:2403.12000, 2024

    Victor Shepardson, Jack Armitage, and Thor Magnusson. Notochord: A flexible probabilistic model for real-time MIDI performance.arXiv preprint arXiv:2403.12000, 2024

  23. [31]

    Learning a latent space of multitrack measures, 2018

    Ian Simon, Adam Roberts, Colin Raffel, Jesse Engel, Curtis Hawthorne, and Douglas Eck. Learning a latent space of multitrack measures, 2018

  24. [32]

    Four treasures of a scholar’s studio, 2026

    Smithsonian National Museum of Asian Art. Four treasures of a scholar’s studio, 2026. Smithsonian Institution, accessed: 2026-02-03

  25. [33]

    Computer assisted melo-rhythmic generation of traditional chinese music from ink brush calligraphy

    Will WW Tang, Stephen Chan, Grace Ngai, and Hong-va Leong. Computer assisted melo-rhythmic generation of traditional chinese music from ink brush calligraphy. InProceedings of the International Conference on New Interfaces for Musical Expression, pages 84–89, 2013

  26. [34]

    Decision of the intergovernmental committee: 4.COM 13.08

    UNESCO. Decision of the intergovernmental committee: 4.COM 13.08. UN- ESCO Intangible Cultural Heritage, 2009. Retrieved February 8, 2026

  27. [35]

    Chinese calligraphy, 2009

    UNESCO Intangible Cultural Heritage. Chinese calligraphy, 2009. Representa- tive List of the Intangible Cultural Heritage of Humanity, accessed: 2026-02-03

  28. [36]

    SongDriver: Real-time music accompaniment generation without logical la- tency nor exposure bias

    Zihao Wang, Qihao Liang, Kejun Zhang, Yuxing Wang, Chen Zhang, Pengfei Yu, Yongsheng Feng, Wenbo Liu, Yikai Wang, Yuntai Bao, and Yiheng Yang. SongDriver: Real-time music accompaniment generation without logical la- tency nor exposure bias. 2022

  29. [37]

    Whole-song hierarchical genera- tion of symbolic music using cascaded diffusion models.arXiv preprint arXiv:2405.09901, 2024

    Ziyu Wang, Lejun Min, and Gus Xia. Whole-song hierarchical genera- tion of symbolic music using cascaded diffusion models.arXiv preprint arXiv:2405.09901, 2024

  30. [38]

    Gesture-driven DDSP synthesis for digitizing the chinese erhu

    Wenqi WU and Hanyu QU. Gesture-driven DDSP synthesis for digitizing the chinese erhu. In Doga Cavdir and Florent Berthaut, editors,Proceedings of the International Conference on New Interfaces for Musical Expression, pages 505–510, Canberra, Australia, June 2025

  31. [39]

    Adaptive accompaniment with ReaLchords

    Yusong Wu, Tim Cooijmans, Kyle Kastner, Adam Roberts, Ian Simon, Alexan- der Scarlatos, Chris Donahue, Cassie Tarakajian, Shayegan Omidshafiei, Aaron Courville, Pablo Samuel Castro, Natasha Jaques, and Cheng-Zhi Anna Huang. Adaptive accompaniment with ReaLchords. InProceedings...

  32. [40]

    Deep music analogy via latent representation disentanglement.arXiv preprint arXiv:1906.03626, 2019

    Ruihan Yang, Dingsu Wang, Ziyu Wang, Tianyao Chen, Junyan Jiang, and Gus Xia. Deep music analogy via latent representation disentanglement.arXiv preprint arXiv:1906.03626, 2019

  33. [41]

    Chinese calligraphy

    Chiang Yee and The Editors of Encyclopaedia Britannica. Chinese calligraphy. Encyclopaedia Britannica, 2009. Accessed: 2026-02-08

  34. [2019]

    Extended abstract (Late-Breaking/Demo, unrefereed)

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.