1. Set up

Same microphone, same distance (about a hand's width), same room, every session. Close the door, turn off fans. Sessions of 30–45 minutes, water nearby. Aim for loud syllables around −10 dBFS on the meter, never red.

Pick a folder (e.g. ~/voice-clone/studio) and every take is written there as it is recorded.

2. Record

3. Make your voice

One command on an Apple-silicon Mac, pointed at a folder of recordings of you talking — the takes above, old screencasts, voice memos; audio or video, sub-folders included:

cd tools/voice-studio/server
./setup.sh                                   # once
conda activate voxcpm
python make_voice.py ~/Recordings/me --name me   # → me.voice.zip

It skips files without audio, other people's voices and already-respoken audio, transcribes, trains (about 20–30 minutes per hour of speech), keeps the best checkpoint and writes me.voice.zip — your voice, versioned and checksummed. Keep it like a password. Then run python voice_server.py and, in the Video Compressor, choose Voice → My voice profile → Load a .voice.zip…

4. Hear your voice

So the first thing you'll notice is that the image is a little dim. Let me show you what happens when we turn on auto-contrast.