Logo image
Back
Timbre space learning for augmentation of musical audio synthesizer interfaces
Dissertation   Open access

Timbre space learning for augmentation of musical audio synthesizer interfaces

Jeff Gregorio
Doctor of Philosophy (Ph.D.), Drexel University
08 Jun 2021
DOI:
https://doi.org/10.17918/00000391
pdf
Gregorio_Jeff_20217.46 MBDownloadView
PDF Dissertation Open Access Open Access (License Unspecified)
m4v
Gregorio_Jeff_2021_Suppl138.59 MBDownloadView
Video (supplemental) Demonstration of novel instrument Open Access Open Access (License Unspecified)

Abstract

Tone color (Music) Human-centered design Music interaction design Musical audio synthesis Machine Learning
Machine learning systems are increasingly being explored for potential applications in the design of creative tools. In particular, musical audio synthesis is one such domain in which these systems can lend to the development of alternative modes of user interaction by means of control spaces based on relevant sonic qualities, rather than traditional synthesis parameters. Although these alternatives effectively lower the barrier to entry to synthesis by obviating the systems-level knowledge required to use traditional parametric synthesizers, we note that with few exceptions, prior work has simply presupposed a general utility in simplified modes of control, or used quantitative efficacy metrics which sacrifice ecological validity for precise evaluation. Meriting further investigation are relevant questions regarding the lack of widespread adoption of such systems, and their relation to the experience levels, creative processes, and aesthetic values that motivate synthesists. Toward answering these questions, this work proposes a novel timbre space learning (TSL) system which uses a generative neural network model to learn mappings from low-dimensional control spaces to synthesis parameters, and an example application which augments a novel parametric synthesizer rather than serving as its primary mode of interaction. A primarily inductive analysis of qualitative and quantitative data from 20 synthesists across a range of experience levels indicates that, when presented with the option of a simplified mode of interaction, users generally give primacy to parameter spaces and seek systems-level understanding of how parameters relate to sonic quality. This work highlights a range of creative uses of TSL augmentation for discovery of new sounds, expressive modulations of parameters, and timbral transpositions. Further, thematic analysis of participant interviews suggests a range of musical aesthetic values and complex relationships to expertise which may have implications for the use of TSL systems.

Metrics

58 File views/ downloads
50 Record Views

Details

Logo image