chances are we'll be able to use the models that process audio to achieve this either way though tbh