Built with Fish Audio. Fish Audio S2 Pro is a large model built for cloning, and it shows: a short, clean recording is enough to get a convincing voice.
Clone
Avocado sends up to four of a voice’s recordings to the model, up to 30 seconds in all. Ten seconds or more clones best. Only clone voices you have permission to use; Avocado asks you to confirm before it makes a clone.
Tags and style
Type tags in square brackets right into the script, like [laugh], [whisper], [sigh] or [pause], or any word that describes the delivery. A style field sets the tone for a whole take.
Dialogue
Cast up to four cloned voices, one per speaker, and generate the conversation as a single take.
Train
Teach S2 Pro a voice from your own recordings. Trained voices appear in your Voice Library with a Trained badge.
What you need
S2 Pro is a big model. It uses about 12 GB of memory while speaking and a little more while training, so it suits Macs with plenty of memory. Avocado lets you choose a faster setting under Performance.
Licence
Avocado shows “Built with Fish Audio” and the licence wherever the model appears. Read the full terms on the Fish Audio S2 Pro model page.