RexaThe short answer
Voice cloning lets an organisation deploy an agent that sounds like a named individual — a founder, a known presenter, a specific support persona. It is technically straightforward and legally the most sensitive thing in a voice stack, because the person whose voice is copied has rights in it and consent has to be captured and provable before the clone is used.
In detail
Two approaches dominate. Few-shot or instant cloning conditions a pretrained multi-speaker model on a short reference sample, producing a usable voice from seconds of audio without any training run. Fine-tuned cloning adapts model weights on a larger, cleaner corpus, which yields better similarity and stability at the cost of time and a real recording session.
The legal position is not uniform and it is moving. Several US states now provide a right of publicity that covers vocal likeness, the EU AI Act imposes transparency duties on synthetic media, and biometric statutes in some jurisdictions treat voiceprints as protected data with their own consent and retention requirements. The reliable posture is to capture explicit, recorded, revocable consent from the speaker and retain the evidence, independent of what any single jurisdiction currently demands.
Cloning is also the most abusable capability in the stack, because a convincing clone of a specific person defeats voice-based identity checks and enables impersonation fraud. Serious platforms gate it behind consent verification rather than treating it as a self-serve upload, and that gate is a feature rather than friction to be removed.
How Rexa handles it
Voice clones on Rexa cannot be finalised without a recorded consent verification. Each clone carries a consent_verification_method of either audio_attestation — a spoken attestation captured from the speaker — or enterprise_manual_review, where the submission enters a human review queue before the clone can be used. The finalisation gate reads that field; a clone that has not cleared its verification path is not dispatchable.
Compliance overviewRelated terms
Text-to-speech
Text-to-speech is the synthesis of natural-sounding spoken audio from written text.
Consent capture
Consent capture is the process of obtaining, recording and retaining evidence that a person agreed to be contacted.
Prosody
Prosody is the pattern of pitch, timing, loudness and rhythm layered over the words of an utterance.
GDPR
The GDPR is the European Union regulation governing the processing of personal data relating to people in the EU and EEA.