Since 2007, the research on the anthropomorphic saxophonist robot at Waseda University aims in understanding the human motor control from an engineering point of view as well as an approach to enable the interaction with musical partners. As a result of our research, last year we have introduced the Waseda Saxophonist Robot No. 1 (WAS-1), composed of 15-DOFs that reproduced the lips (1-DOF), tonguing (1-DOF), oral cavity, lungs (2-DOF) and fingers (11-DOFs). However, even that the mouth mechanism of WAS-1 was useful in order to adjust the pitch of the saxophone sound, the range of sound pressure was too narrow. Thus, no dynamic effects of the sound can be reproduced (i.e. crescendo and decrescendo). Moreover, the finger mechanism was designed only to play from C3-C#5. On the other hand, a cascade feedback control system has been implemented in the WAS-1; however, a considerable delay in the attack time to reach the desired air pressure was detected. Therefore, in this paper, the Waseda Saxophone Robot No. 2 (WAS-2) which is composed by 22-DOFs is detailed. The lip mechanism of WAS-2 has been designed with 3-DOFs to control the motion of the lower, upper and sideway lips. In addition, a human-like hand (16 DOF-s) has been designed to enable to play all the keys of the instrument. Regarding the improvement of the control system, a feed-forward control system with dead-time compensation has been implemented to assure the accurate control of the air pressure. A set of experiments were carried out to verify the mechanical design improvements and the dynamic response of the air pressure. As a result, the range of sound pressure has been increased and the proposed control system improved the dynamic response of the air pressure control.