AudioGen seems really fascinating. I have some dumb questions. While the datasets used for training AudioGen aren't available, is there any kind of list where

AudioGen seems really fascinating. I have some dumb questions. While the datasets used for training AudioGen aren't available, is there any kind of list where one can review the tags or descriptions of the sounds on which the model was trained? Otherwise how do you know what kinds of sounds you can reasonably expect AudioGen to be capable of generating? And what happens if you request a sound which is too obscure or something not found in the dataset? What are AudioGen's capabilities regarding spatial positioning? First example: can it generate a siren that starts in front and moves left to right and complete a full circle around the listener? Second example: can it do the same siren but on the Y axis, so it start at the front, it goes over the listener and then it goes under them to complete the circle?