Replace voice-clone audio and its transcript as one unit

When a voice-cloning service uses transcript-conditioned reference audio, replace the audio and its matching transcript together. A valid new recording paired with the previous recording's text can corrupt synthesis even while health checks, latency, and request completion remain normal. Verify the words spoken in the new reference, restart or rebuild the cached clone prompt, and test actual speech before routing traffic. Matching model weights alone does not validate this pair. This applies to transcript-conditioned cloning; speaker-only conditioning has a different contract. Applies when: A migration or voice swap changes the recording used to condition speech. Keywords: voice cloning reference audio transcript mismatch migration conditioning