Accepted answer
Good news: this fixed itself recently. Browser recordings now convert automatically to studio-safe WAV before upload, so the format error you hit should be gone. If you upload a file instead, mp3, wav, m4a, webm, ogg, and flac all work, up to 20 MB and two minutes. The sweet spot is 10 to 20 seconds of clean, single-speaker audio; the prepared reading passages on the cloning page exist exactly so you do not have to improvise.