Decoder. plain-English AI glossary

Voice Cloning

● Core

Creating a synthetic voice that sounds like a specific person from a few audio samples—read text in your voice.

Think of it like

Recording yourself saying different phonemes then auto-combining them to say anything.

Example

30 seconds of your voice; AI generates "I will call you later" in your accent, tone, and speaking style.

How it actually works

Speaker encoder learns embeddings from reference audio (speaker verification). TTS conditioned on speaker embedding. Training requires 1-10 hours of target speaker data for good quality; few-shot with 10 seconds possible but lower quality. Can be misused for deepfakes; watermarking and detection critical.

For product teams

Personalization and accessibility; also raises consent and fraud concerns.

For engineers

Speaker encoder extracts speaker embedding from reference; TTS conditioned on embedding; attention learns target speaker prosody.

Related

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open DecoderAdd to Chrome