Select an emotion and speaker, then specify your text to generate!
Higher values make output more random.
Limits sampling to the K most probable tokens.