Voice Cloning

Voice cloning lets a TTS engine produce new speech that sounds like a specific person, given just a short reference recording. ElevenLabs Instant Voice Cloning works from 1 to 3 minutes of audio; their Professional Voice Cloning uses 30+ minutes for higher quality. The technique is now indistinguishable from real human speech in short utterances. Common production uses: branded voice for a company's voice agent, restoring a known voice for accessibility, or letting a power user record their own voice for personal AI apps. Most providers require explicit consent verification (you record a verification phrase) before they let you clone.

Voice cloning has serious safety implications. The same technology that makes accessibility apps possible enables convincing audio scams. Every reputable provider now requires consent verification for non-public voices and watermarks generated audio. Some jurisdictions (EU AI Act, California voice-likeness laws) require explicit disclosure when synthesized voice is used. For business voice agents, the practical guidance: clone the voice of someone who has explicitly consented, never of a celebrity or public figure without permission, and document the consent. Use a clearly different voice from any real employee to avoid identity confusion.

Clone a specific voice when brand consistency or accessibility requires it. Use a stock voice for general-purpose agents to avoid the consent and disclosure burden.

• ElevenLabs voice cloning docs
• Cartesia voice cloning

Related Terms