Describe the voice you can hear in your head.

A useful voice description has three parts

Start with the speaker’s general character, add the vocal texture, and finish with the intended setting. For example: “A thoughtful young adult with a clear, gentle tone, speaking like a late-night radio host.” Avoid imitating a real person without permission or requesting a deceptive impersonation.

• Character: warm, playful, grounded, elegant, energetic
• Texture: clear, husky, soft, resonant, airy
• Context: documentary, bedtime story, product guide, conversation

Identity and performance are different

The voice description defines the broad sound. The performance direction explains how this particular passage should be delivered. Keeping them separate makes a voice reusable across calm introductions, excited announcements, and reflective endings.

A fixed seed improves repeatability, but generated speech can still vary with text, punctuation, and other controls. Treat the seed as part of a complete recipe rather than a universal voice identifier.

Design responsibly

Create fictional or properly authorized voices. Do not design speech to mislead listeners about who is speaking, fabricate endorsements, bypass identity checks, or cause harm. When a listener could reasonably mistake synthetic speech for a real person, disclose that the audio is AI-generated.