The paper doesn't even cover the intermediate neuronal stages in the brainstem, which are extremely important for localization and detecting of different featur

The paper doesn't even cover the intermediate neuronal stages in the brainstem, which are extremely important for localization and detecting of different features of sound - in fact several of those are well characterized in the brainstem. Your brainstem can detect level and timing differences for a sound for incoming sounds from each ear (interaural level and interaural time differences), depending on the frequency of the sound, and uses that for localization. As for the improbability - it makes perfect sense when you think of each layer being some sort of improvement on the previous layer(s) rather than a well-engineered preplanned system.