Text this: Knowledge enhancement for speech emotion recognition via multi-level acoustic feature