図 5.1: Block diagram of mora-timed speech rhythm training
G.Kawai and K.Hirose (1998) investigated the Tokushuhaku duration and its intelligibility perceived by native speakers[36]. To clarify its exact durations perceived by native speakers, they did perceptual experiments. Minimal pairs of actual Japanese words of the presence or absence of Tokushuhaku were prepared, e.g., kado (a corner) and kaado (a card), nasu (an eggplant) and naasu (a nurse), kuro (black) and kuuro (an air route), etc. Next, for each pairs of word, 13 synthesized words were generated along with inserting a pause according to its duration between Tokushuhaku and non-Tokushuhaku. Vowel durations were adjusted at roughly constant ratios. All different varieties of each word were played twice in random order. Twelve native speakers of Japanese were asked to categorize the words as Tokushuhaku or non-Tokushuhaku word. They showed that there was almost perfect agreement with regard to short and long durations, whereas mid-range durations were judged as ambiguous. Their discrimination curves closely matched normal distributions. However, they succeeded just in establishing intelligibility functions for each word, not solving Tokushuhaku intelligibility itself.
図 5.1: Block diagram of mora-timed speech rhythm training
