A binaural recording and reproduction system using a dummy head microphone (DHM) has been used in various applications such as virtual reality systems, transaural reproduction systems for entertainment, and evaluations of many sound fields. However, the size and weight of the DHM have been inconvenient for many practical uses. An alternative to an ordinary DHM was proposed which consisted of a microphone array and a neural network controller. Thanks to the electronic configuration of the new DHM, many advantages are expected. For example, not only will it include a downsized microphone, but also several advantages, including multiple HRTF configurations, could be realized. In this paper, an efficient algorithm is used in the learning process of the neural network. The results of subjective evaluation for perceived directions compared to those of the ordinary DHM will be shown.