Partial devoicing of voiced geminate stops in Tokyo Japanese

Tokyo Japanese has a constraint against voiced geminate stops in its native lexicon. The present study investigates whether recently introduced word-medial voiced geminate stops [C1V1C(C)2V2] are differentiated from voiceless geminates and voiced singletons in terms of duration, voicing during closure, and spectral moments of stop release bursts. The findings suggest that the voiceless and voiced singleton stops were clearly differentiated by C2 duration. In contrast, C2 duration of the voiceless and voiced geminate stops was not significantly different. The devoicing of the word-medial stops was not only observed in voiced geminates, but voiced singletons also showed devoicing. The duration of the preceding vowel (V1) distinguished the voicing contrast in both singleton and geminate stops. The first four spectral moments of C2 stop release bursts did not distinguish the length and voicing contrasts in stops. These results indicate that, although word-medial voiced geminate stops are fully or partially devoiced, the Tokyo Japanese speakers lengthen the preceding vowels (V1) to maintain a voicing contrast. Production patterns of the voiced geminates are considered in relation to marginal or intermediate phonological contrast.