This study demonstrates auditory human fMRI conducted at 9.4T field strength and submillimeter resolution for the first time. Tonotopic maps were measured robustly and reliably. Further, cortical regions with preference for natural sound categories were delineated. We generated ripple control sounds that closely match low level acoustical properties of natural sounds in four natural sound categories, such that the original category is not recognizable. We show that, in areas preferring speech sounds over other natural sounds, ripple control sounds of speech elicit stronger responses than ripple control sounds of non-speech. This indicates tuning to the low-level acoustical properties of speech.