NeuMu

NeuMu aims to give artists a more accessible interface for controlling RAVE-based generative audio. It approaches control through artistic parameters associated with perceived sound qualities, connecting a desired change in sound with guidance in a model’s latent representation. The project explores how artists can shape generated sound through listening, adjustment, and an understanding of how a particular model responds.

The documented interface includes descriptors such as loudness, pitch, brightness, noisiness, roughness, and activity. Their availability depends on the model and the checks carried out on its controls. These descriptors guide a latent trajectory before RAVE decodes it into audio. Their effects can interact, and the relationship between a descriptor and the resulting sound remains specific to the model.

An existing example compares two saved outputs from the same water-texture model and source trajectory. One is a raw RAVE replay; the other uses guidance, with Roughness selected during part of the excerpt. The paired spectrograms align the sounds with the recorded control interval, providing material for examining how the output changes. Other factors such as decoding and state also shape the result.

INTERPRET draws on many of NeuMu’s functions and materials within its dance-based audiovisual system. This connects the interface’s development with an artistic project in which generative audio forms part of an evolving exchange between movement, sound, and visuals.

A and B document different models and controllers.

The documented controls remain model-dependent. Listening and the saved-output comparison provide material for examining changes in sound, but decoding, state initialization, feedback, and transitions can also affect the result.