Cloning is async: rapid needs 3+ recordings, professional needs 20+, and the voice only works once training shows finished. Use a callback_uri instead of polling.

Context: Official docs (Resemble voice clone overview): rapid cloning needs at least 3 recordings (about 10 seconds of audio) and professional cloning needs 20 or more (about 10 minutes). Training is asynchronous either way, and the voice is only usable once training reaches finished status. Set a callback_uri so you get told when it completes or when the dataset has quality problems, instead of polling.