customize a classifier

Start from a published base model, tune the vocabulary to your task, then bake a fresh .rivlet binary + cross-platform bundle. ← back to /inference

Customizing from voice-numbers by Rivlet. The bake will deliver a per-user signed model to your library.

Vocabulary editor

Existing classes — keep or remove

0 / 0 kept

Removing prunes the corresponding regions from the substrate at bake time — real pruning, not an output mask.

(base has no preset classes — fully custom from scratch)

Add custom words / sounds

0 / 10 added

New classes get fresh substrate regions grown into the base model — actual substrate-level growth, not an adapter hack. 5+ samples per word is enough; transfer from the base does the rest.