Skip to main content
OT

Voice cloning

Also known as：voice replication · speaker cloning · custom voice

Voice cloning is the technique of reproducing a specific speaker from a short recording sample. In cross-border live streaming it makes the translated audio still sound like the host, preserving the personal continuity the format depends on.

Same-language versus cross-lingual

Same-language cloning — a Chinese voice speaking Chinese — is technically mature.

Cross-lingual cloning — a Chinese voice speaking English, Indonesian or Vietnamese — varies enormously, and generally degrades as the target language becomes less common. Always audition your own voice in your target language. Vendor demos are selected for the combinations that work best.

Costs

  • Higher latency. Cloned synthesis is generally slower than stock synthesis, and the overhead is proportionally largest on short utterances.
  • Flat affect. Live delivery has a wide emotional range; most cloned voices hold one fairly constant register.
  • Narrower language coverage. Fewer target languages are supported than with stock voices.

Cloning your own voice requires explicit, documented consent from that person. Cloning anyone else — employees, partner hosts, public figures — is infringement without authorisation. Article 1023 of the Chinese Civil Code applies the protections governing portrait rights to a natural person’s voice.

The trap worth knowing: permission to use a recording is not permission to clone the voice in it. That is precisely what the Beijing Internet Court decided in China’s first personality-rights case over an AI-generated voice, (2023) Jing 0491 Min Chu No. 12142 — the voice actor had licensed the use of the recordings, not their conversion into an AI voice by a third party. Identifiability is the test: if an ordinary listener associates the synthesised voice with that person, it falls within the protected interest.

Even with your own voice, using synthesis to make commercial claims about price, efficacy or after-sales terms that you never made does not shift liability. The safe operating rule is to treat a clone strictly as a translation stand-in for content the person actually said.

Related terms