For a good Cartesia instant clone, record 10-60 seconds of one speaker in a quiet room, speaking naturally in the mood you want the voice to keep. Clone in the speaker's own language first and add other languages via the Add Voice Accents API afterward. If the voice has a rare accent or distinctive qualities, use a Pro Voice Clone with 30+ minutes of audio instead.

Context: Official Cartesia docs (Instant Voice Clone guide): documents the recording gotchas that make clones sound off. The clone picks up the mood of your clip, so a cheerful take produces a voice that sounds cheerful in everything it says; record in the mood you want the final voice to have. Record one speaker in a quiet room with no music or echo, and speak naturally rather than announcing. Each clone is built from one language, so record in a language the speaker actually speaks, then add more languages with the Add Voice Accents API. Ten seconds is enough to start; up to 60 seconds better retains the accent.