Minifusion

Minifusion is an audio plugin for sound-based transformations of musical instruments, that allows to shape the sound of an incoming audio signal in the likeness of other musical instruments, while running in real time and low latency. This project is motivated by the difficulty of musicians playing non-keyboard musical instruments (e.g. guitarists, vocalists, percussionists) to access to a wide variety of sounds. This is because many production tasks such as music sketching, composition in Digital Audio Workstations, and sound design, require an interaction paradigm centered around keyboard interfaces and MIDI, which is inherently incompatible with the embodied, continuous, and contextual mechanisms of interaction with non-keyboard instruments. The plugin and neural network design have been in continuous development while in collaboration with a few musicians and a composer, and to date, affords transformations of different audio sources such as guitar and voice, into other specific instruments (like a rompler but without MIDI!). It is also possible to train custom models. Future development includes expanding the sound shaping possibilities.

This plugin implements two main features to afford natural control for instrumentalists, stemming from research conducted at the lab: (1) the use of continuous representations with space for ambiguity, that holds potential for meaning making when playing live, and (2), such representations are learn in low latency autoencoders that support low latency and jitter, for intimate instrumental control.

This plugin (and its older versions) have been presented at different venues such as Sonar 2025, NIME and WASPAA, showing potential interest, has registered 100 downloads to date, and has been employed for live use by a few musicians, including a composer preparing a piece for early 2026.