glossary
verified July 2026
Voice cloning reference audio
The reference is the short clip of a real speaker that a zero-shot cloning system conditions on at request time. Everything about the clone — identity, accent, pacing — is inherited from it, which makes the reference the most consequential audio file in the pipeline.
01
The working spec
Fig. — What a working reference is
5 – 10 s
the working range — one speaker, talking naturally
~1.5 MB
the WAV cap, around 30 seconds of audio
1 room
clean beats long: a quiet room and a single voice
Consent is a requirement, not a courtesy — the speaker in the clip has agreed to be cloned.
02
How the reference travels
There is no upload step or voice-management API here: the clip rides inside the synthesis request. References are fingerprinted and cached, so the first request pays the cloning work and every reuse of the same clip skips it — the request shape itself is on the voice cloning capability sheet.
03
What happens to it afterward
Nothing. Reference audio never enters a training set, and the synthesized output it produces carries an inaudible provenance watermark applied at generation — the output is traceable as synthetic, and the reference stays yours.
See also
Related sheets.
glossary
10 s
Instant voice cloning
Instant cloning builds a voice from a ten-second reference at request time — no training job. What zero-shot means, and how identity survives translation.
capability
0
Voice cloning from ten seconds, in the request
Zero-shot voice cloning: a ten-second reference rides inside each request, no training job, and the identity holds across 23 languages.
capability
never
Provenance and privacy, applied at generation
The policy surface of the engine: an inaudible watermark on every sample, reference audio that never trains, and per-key audit trails at GET /v1/usage.
Every term on this page is measurable on a live call — watch the readout while your own script synthesizes.
See the measured figures live