Key points are not available for this paper at this time.
Machine learning (ML) has been used to create mappings for digital musical instruments for over twenty-five years, and numerous ML toolkits have been developed for the NIME community. However, little published work has studied how ML has been used in sustained instrument building and performance practices. This paper examines the experiences of instrument builder and performer Laetitia Sonami, who has been using ML to build and refine her Spring Spyre instrument since 2012. Using Sonami's current practice as a case study, this paper explores the utility, opportunities, and challenges involved in using ML in practice over many years. This paper also reports the perspective of Rebecca Fiebrink, the creator of the Wekinator ML tool used by Sonami, revealing how her work with Sonami has led to changes to the software and to her teaching. This paper thus contributes a deeper understanding of the value of ML for NIME practitioners, and it can inform design considerations for future ML toolkits as well as NIME pedagogy. Further, it provides new perspectives on familiar NIME conversations about mapping strategies, expressivity, and control, informed by a dedicated practice over many years.
Fiebrink et al. (Mon,) studied this question.