A nn~ model's methods each define a distinct processing pipeline with its own inlet/outlet count

A neural model loaded in nn~ can expose multiple named methods — for example a RAVE model typically has encode, decode, and forward. Each method defines a different audio processing pipeline and can have a different number of inlets and outlets. The method is specified as the third argument to the nn~ object. The forward method is the default when no method is given. encode takes audio and produces latent dimensions; decode takes latent signals and produces audio; forward chains both into a timbre-transfer path. Understanding methods is essential to latent-space manipulation in a live patch.

Examples

In Max: nn~ rave.ts forward — one audio inlet, one audio outlet (timbre transfer). nn~ rave.ts encode — one audio inlet, 16 signal outlets (one per latent dimension). nn~ rave.ts decode — 16 latent inlets, one audio outlet.

Assessment

Explain why encode and decode have different inlet/outlet counts from forward. Design a patch that manipulates the latent space between encode and decode, and identify what signals flow on each connection.