Reference audio must be 1-15 seconds for generation. Upload longer audio and use the scissors tool to trim it; 3-10 seconds of clear speech is recommended.
Upload or record reference audio to transcribe it automatically. The transcript is optional.