速さ感

高さ感

年齢推定

naturalness

perceived speed

perceived pitch

age estimation

Many previous studies that used synthetic speech stimuli have artificially manipulated duration to investigate speech perception and speaker inference. These studies have often run experiments with a small number of duration conditions, e.g. slow, normal, fast, because of other constraints. However, the choice of duration conditions is often arbitrary; range of duration that feels like there has been a change in speaking rate while still sounding natural has not been directly investigated. The purpose of the present study is to determine the range of duration in synthesized speech that is wide enough yet natural, and to explore speech perception and speaker inference within such a range. To achieve this goal, duration of recordings of a typical first-time greeting in Japanese was manipulated in 19 steps. Participants listened to the stimuli and evaluated the perceived speed, pitch, and naturalness of the utterance, along with inference of age and voice impression. Results revealed that a range of duration that feels like there has been a change in speaking rate but is not unnatural was 63.51% - 128.45% of the original rate (10.23 – 5.28 moras per second). Within this range of duration, inference of voice impression significantly changed. These results provide an empirical basis for selecting appropriate ranges of duration variation for investigating aspects of speech perception and speaker inference.

501 学内論文 ＞ 紀要 ＞ 大学院紀要＝Bulletin of graduate studies ＞ 89