2026
Peachey, Matthew
Investigating Continuous Interactions for generative Tools in the Context of Musical Creativity PhD Thesis
Dalhousie University, 2026.
Abstract | BibTeX | Tags: continuous interaction, continuous-interaction-and-machine-learning, control mapping, creativity support tools, digital musical instruments, machine learning, music, sound & music computing
@phdthesis{Peachey2026b,
title = {Investigating Continuous Interactions for generative Tools in the Context of Musical Creativity},
author = {Matthew Peachey},
year = {2026},
date = {2026-10-19},
urldate = {2026-10-19},
school = {Dalhousie University},
abstract = {This thesis investigates continuous interaction as a design principle for Artificial Intelligence (AI)-powered music creativity support tools. Discrete text prompting, the de facto standard for current generative AI systems, can limit musicians’ sense of artistic agency and creative ownership. Continuous interaction instead can enable fine-grained, real-time control over the generative process, helping musicians retain a sense of authorship over the generated output.
This thesis presents three studies that together examine how continuous interaction, integrated into real-time interfaces, can better support musicians’ creative workflows. First, I derived low-dimensional latent representations from a Variational Autoencoder with an intentionally small latent space, integrated them into a real-time GUI, and evaluated them in a mixed-methods study with both novice and experienced synthesizer users (n = 18). Second, I interviewed experienced musicians (n = 10) about their hesitations toward generative AI; their concerns centered on a lack of compelling tools and workflows, further motivating continuous interaction as a design preference. Finally, I developed a framework for Aging Digital Musical Instruments (AgingDMIs), a novel class of instruments that adapt to a user’s play style over time. I then implemented a hardware prototype instantiating one point in the AgingDMI design space, and evaluated it longitudinally with experienced musicians (n= 4) for twelve recurring sessions over several weeks.
The findings presented in this thesis suggest that integrating continuous interaction into generative AI-powered creativity support tools increases musicians’ artistic agency and creative ownership and fits better within their existing creative workflows. I further contribute a validated
design framework for AgingDMIs, a novel class of adaptive and personalized digital musical instruments, and a hardware prototype that together demonstrate the principles of that framework in practice.},
keywords = {continuous interaction, continuous-interaction-and-machine-learning, control mapping, creativity support tools, digital musical instruments, machine learning, music, sound & music computing},
pubstate = {published},
tppubtype = {phdthesis}
}
This thesis investigates continuous interaction as a design principle for Artificial Intelligence (AI)-powered music creativity support tools. Discrete text prompting, the de facto standard for current generative AI systems, can limit musicians’ sense of artistic agency and creative ownership. Continuous interaction instead can enable fine-grained, real-time control over the generative process, helping musicians retain a sense of authorship over the generated output.
This thesis presents three studies that together examine how continuous interaction, integrated into real-time interfaces, can better support musicians’ creative workflows. First, I derived low-dimensional latent representations from a Variational Autoencoder with an intentionally small latent space, integrated them into a real-time GUI, and evaluated them in a mixed-methods study with both novice and experienced synthesizer users (n = 18). Second, I interviewed experienced musicians (n = 10) about their hesitations toward generative AI; their concerns centered on a lack of compelling tools and workflows, further motivating continuous interaction as a design preference. Finally, I developed a framework for Aging Digital Musical Instruments (AgingDMIs), a novel class of instruments that adapt to a user’s play style over time. I then implemented a hardware prototype instantiating one point in the AgingDMI design space, and evaluated it longitudinally with experienced musicians (n= 4) for twelve recurring sessions over several weeks.
The findings presented in this thesis suggest that integrating continuous interaction into generative AI-powered creativity support tools increases musicians’ artistic agency and creative ownership and fits better within their existing creative workflows. I further contribute a validated
design framework for AgingDMIs, a novel class of adaptive and personalized digital musical instruments, and a hardware prototype that together demonstrate the principles of that framework in practice.
This thesis presents three studies that together examine how continuous interaction, integrated into real-time interfaces, can better support musicians’ creative workflows. First, I derived low-dimensional latent representations from a Variational Autoencoder with an intentionally small latent space, integrated them into a real-time GUI, and evaluated them in a mixed-methods study with both novice and experienced synthesizer users (n = 18). Second, I interviewed experienced musicians (n = 10) about their hesitations toward generative AI; their concerns centered on a lack of compelling tools and workflows, further motivating continuous interaction as a design preference. Finally, I developed a framework for Aging Digital Musical Instruments (AgingDMIs), a novel class of instruments that adapt to a user’s play style over time. I then implemented a hardware prototype instantiating one point in the AgingDMI design space, and evaluated it longitudinally with experienced musicians (n= 4) for twelve recurring sessions over several weeks.
The findings presented in this thesis suggest that integrating continuous interaction into generative AI-powered creativity support tools increases musicians’ artistic agency and creative ownership and fits better within their existing creative workflows. I further contribute a validated
design framework for AgingDMIs, a novel class of adaptive and personalized digital musical instruments, and a hardware prototype that together demonstrate the principles of that framework in practice.
Shah, Arsh
RowSim: Designing and Evaluating Ambient Interaction in Mixed-Reality Rowing Masters Thesis
Dalhousie University, 2026.
Abstract | BibTeX | Tags: ambient display, control mapping, immersive visualization, IMU, projection mapping, rowing, sensor fusion, simulation
@mastersthesis{Shah2026,
title = {RowSim: Designing and Evaluating Ambient Interaction in Mixed-Reality Rowing},
author = {Arsh Shah},
year = {2026},
date = {2026-08-31},
urldate = {2026-08-31},
school = {Dalhousie University},
abstract = {A rowing ergometer reproduces the effort of rowing while removing its perceptual accom-
paniment, filling the gap with a console that must be read. Ambient-display research has
argued for three decades that information can occupy the periphery, but almost entirely for
people at rest.
RowSim augments an ergometer with a floor-projected fluid simulation driven by the
rower’s movement. Effort scales the wake’s brightness and how hard it pushes the fluid;
nothing is a number or recoverable as a value. Seventeen rowers, twelve casual and five
elite, rowed an open-ended ten-minute session with the console covered, then answered
questionnaires and an interview.
The groups differed by a factor of 3.7 in mean power, with all seven physical measures
separating them; no engagement measure did. Two participants independently reported that
the flow did not read as their own, because no action could stop it: restoring optic flow is a
coupling problem, not a rendering one.
The contribution is a working system and evidence that a peripheral, non-symbolic
display remains usable during exertion, together with a perceptual account of why the
missing accompaniment matters and of the one condition the display failed to meet.},
keywords = {ambient display, control mapping, immersive visualization, IMU, projection mapping, rowing, sensor fusion, simulation},
pubstate = {published},
tppubtype = {mastersthesis}
}
A rowing ergometer reproduces the effort of rowing while removing its perceptual accom-
paniment, filling the gap with a console that must be read. Ambient-display research has
argued for three decades that information can occupy the periphery, but almost entirely for
people at rest.
RowSim augments an ergometer with a floor-projected fluid simulation driven by the
rower’s movement. Effort scales the wake’s brightness and how hard it pushes the fluid;
nothing is a number or recoverable as a value. Seventeen rowers, twelve casual and five
elite, rowed an open-ended ten-minute session with the console covered, then answered
questionnaires and an interview.
The groups differed by a factor of 3.7 in mean power, with all seven physical measures
separating them; no engagement measure did. Two participants independently reported that
the flow did not read as their own, because no action could stop it: restoring optic flow is a
coupling problem, not a rendering one.
The contribution is a working system and evidence that a peripheral, non-symbolic
display remains usable during exertion, together with a perceptual account of why the
missing accompaniment matters and of the one condition the display failed to meet.
paniment, filling the gap with a console that must be read. Ambient-display research has
argued for three decades that information can occupy the periphery, but almost entirely for
people at rest.
RowSim augments an ergometer with a floor-projected fluid simulation driven by the
rower’s movement. Effort scales the wake’s brightness and how hard it pushes the fluid;
nothing is a number or recoverable as a value. Seventeen rowers, twelve casual and five
elite, rowed an open-ended ten-minute session with the console covered, then answered
questionnaires and an interview.
The groups differed by a factor of 3.7 in mean power, with all seven physical measures
separating them; no engagement measure did. Two participants independently reported that
the flow did not read as their own, because no action could stop it: restoring optic flow is a
coupling problem, not a rendering one.
The contribution is a working system and evidence that a peripheral, non-symbolic
display remains usable during exertion, together with a perceptual account of why the
missing accompaniment matters and of the one condition the display failed to meet.
