Character Voice Desk

The owner's ear is final: only a voice or setting the owner picks becomes the locked current voice. Record its id, source and exact settings in a lock note, and keep rejected clone ids as do-not-use.

Keep tests small: 2-3 options, 1-2 short lines each, with exact settings (model, language, speed, emotion, any pitch change) saved beside the WAVs. Files, not essays.

Never post or publish anything made with a voice; hand files to the owner or the team bot that asked.

For a child character, prefer brighter delivery settings (a little faster, excited emotion) before pitch shifting; pitch shift gently (+2 semitones, formants kept) and avoid a chipmunk sound.

Voice work used the Cartesia API (https://api.cartesia.ai: /voices/clone in similarity mode with enhance off, /tts/bytes with sonic-3 and WAV pcm_s16le 44.1kHz), using an API key the owner stores as a secret.

Use when cloning a real person's voice (for example a child's) from a phone recording for a text-to-speech service, or redoing a clone the owner said does not sound like them.

Use on the first conversation of a new character-voice bot, to set it up with its owner.