Laetitia Sonami’s Spring Spyre begins with a metal ring, three springs and pickups salvaged from spring reverberation units. Their vibrations become control signals for electronic sound. Sonami began developing the instrument in 2011.[1]
Its software includes Wekinator, Rebecca Fiebrink’s free, open-source tool for learning relationships from demonstrations. A creator supplies examples pairing an input with a desired response. The trained model then turns incoming signals into output values. Wekinator connects to other software through Open Sound Control, a protocol used to exchange control messages.[2]
Training the relationship
Imagine making a much simpler instrument: a hand tilt controls a synthesiser. You demonstrate a small tilt beside a quiet, dark tone, then a larger tilt beside a brighter, louder one. The interesting design question is what should happen between those examples. Should the sound change smoothly? Should most of the movement leave it nearly unchanged, with a dramatic transformation near the end? Different answers would give the same physical object a different playing character. That is an illustrative design exercise, not a description of Sonami’s training examples.
In their first-hand 2020 paper, Sonami and Fiebrink describe training the Spring Spyre on performed signals paired with synthesis settings. During performance, Sonami can hold selected input values steady while others keep changing, or hold them all. This lets her remain within a narrower region of sound.[3]
The performer chooses when to move through a sound and when to linger. An interesting response becomes something to explore, with listening helping to guide the controls.[3]
The collaboration also influenced Wekinator’s interface: selecting which inputs affect particular outputs and improving the handling of training examples became practical design priorities.[3]
For an interactive installation, try writing the intended relationship before choosing the model: a visitor’s distance controls the spread of a sound, while their speed controls how abruptly it changes. Then ask whether they should be able to find a stable point and return to it. Those decisions specify an experience a person can learn to play.
Sources & further reading
- Spring Spyre Laetitia Sonami · Accessed 20 Oct 2026
- Wekinator Rebecca Fiebrink · Accessed 20 Oct 2026
- Reflections on Eight Years of Instrument Creation with Machine Learning Fiebrink and Sonami, NIME 2020 · Accessed 20 Oct 2026