Zero-shot voice cloning
Also known as: instant cloning, few-shot cloning
Zero-shot voice cloning reproduces a voice from a few seconds of reference audio, with no separate training step. The model has already learned how voices work in general, so it adapts to a new one on the fly, trading some control for speed.
Why it matters
Zero-shot cloning removes the biggest barrier to a clone: the training step. Instead of recording a dataset and waiting, you hand the model a short reference and it speaks in that voice immediately. That makes it practical to try many voices, or to clone from whatever short clip you already have.
The trade is control and consistency. Because nothing is fitted to your voice specifically, quality varies more from run to run, and long or highly expressive passages are where the difference from a trained profile shows.
In practice
Give the model a few seconds of clean, dry reference audio: no music, no reverb, one speaker. Expect a strong result on short lines and more variance across a long script. If you need the same voice to sound identical every time, a fitted profile beats zero-shot.