Re-enroll an existing voice: pick the best snippet of a new recording and swap it in.

The same selection as POST /voices/assisted, applied to a voice that already exists. Its name, tags and id are kept; its reference (transcript, codes and markers at every speed) is replaced in one write, so a rejected recording leaves the voice exactly as it was.

Args: voice_id: The voice to update. lang_code: Language of the new recording. Defaults to the voice's own. script: The text the speaker read. Defaults to the built-in script for the language. preview_text: Line to speak back with the updated voice; an empty string skips it. fast: Use our own enhancer, as in POST /voices/assisted. candidates: Re-enroll the whole candidate set voice_id belongs to, as POST /voices/assisted?candidates=true made it: a candidate whose name comes again keeps its voice id and is otherwise rewritten as a new candidate (unchosen, with the new reference), a new one is added, and one the new recording does not give is deleted. The answer is the new set, as POST /voices/assisted gives it. A voice enrolled without candidates is refused. voice_file: The new enrollment recording, the whole script in one take, 15-60 seconds.