×
Fifteen seconds of audio, one line in your terminal, and Claude speaks in your voice. No microphone, no retakes, no subscription.
You are not recording any more.
You are directing.
Run this once in the project where you want Claude to write Fish Audio code. It drops both skills into your agent, with symlinks for Claude Code and Cursor.
npx skills add https://docs.fish.audio
You can also reach it from Google: search fish audio coding agents and open the first result.
Make a free Fish Audio account, create a key in your dashboard, copy it, and paste it back into Claude when it asks. No card, no plan.
Open Fish Audiofish.audio/app/api-keys
Treat it like a password. Never show it on camera or paste it in a screenshot.
Drop fifteen seconds of yourself talking into the chat and type three words. Claude trains it on S2.1 Pro, which is the free model.
Clone my voice
This is the part almost nobody uses. Write the tag inline, anywhere in the script, and the model performs it. You are writing stage directions, not typing text.
[warm] This is a test of my cloned voice. [amused] If this sounds like me, [measured] the setup worked.
If you build something with it, send it to me. I want to hear how far this goes.
Start free on Fish Audio Message me on Instagram @marcelgrauu