リアルタイム音高制御に用いる上肢動作の比較評価

抄録

A real-time speech synthesizer enables users to control pitch, loudness, and phonemic information of the output speech using their body motions in real time. To investigate optimal motions for the synthesizer, pitch control experiments were carried out. A subject controls hand height, shoulder angle, or elbow angle to determine pitch of the matching tone, and is required to match with the pitch of the presented target tone. The pitch errors were evaluated in two sensory feedback conditions. One is a full-feedback conditions in which auditory, kinesthetic, and visual feedbacks are available. The other is a limited feedback condition where visual feedback is unavailable. The results indicated that subjects could control hand height with 4.2 mm accuracy, shoulder angle with 0.85- degree accuracy, and elbow angle with 1.8-degree accuracy in the full-feedback condition. The visual feedback was found to be very effective which could reduce 42.1 to 64.1 % pitch errors on the average. In the limited feedback condition, the mean pitch errors were 0.465, 0.546, and 0.410 semitone for hand, shoulder, and elbow controls, respectively. The elbow angle was thus concluded to be the most effective motion to control pitch for the speech synthesizer.