We reproduced the feedback mechanism of human speech for an anthropomorphic talking robot WT-5 (Waseda Talker No. 5). The Waseda Talker series mimics human speech organs such as the vocal cords, tongue and lips, and WT-5 could produce various vowels and consonant sounds. We elucidate the mechanisms of human speech and speech acquisition using this robot. In human speech acquisition, humans build phoneme categories through mimic speaking. In mimic speaking, infants reproduce adult voices by a feedback mechanism. We have developed a vowel mimic speaking mechanism using auditory feedback that optimized sound pressure and the first and second formants (F1, F2). We also developed consonant mimic speaking using sensory information such as tactile and pressure sensors. Moreover, we developed consonant mimic speaking by auditory feedback. In the feedback mechanism, we constructed acoustic features from MFCCs (Mel-Frequency Cepstral Coefficients).