A Study on Final Devoicing in Russian through Quantitative Analysis

This study analyzes final devoicing in Russian using quantitative methods. It is well-known that, in Russian voiceless consonants and voiced consonants that undergo voiceless assimilation (e.g., /pod/ vs. /pot/) are not acoustically identical in terms of the preceding vowel length, indicating the occurrence of incomplete neutralization. However, previous studies have relied on small-scale, tightly controlled manual data. This study attempts a corpus-based linguistic analysis of vowel length, one of the main cues to incomplete neutralization, by using a large speech corpus(RUSLAN) and the forced aligner tool Montreal Forced Aligner(MFA). The results show that, for all voiced-voiceless pairs, the vowel preceding the voiceless-assimilated voiced consonant tends to be longer. However, exceptions were observed in certain consonant pairs, which deviate from the initial hypothesis. These findings suggest the potential for phonetic research using large, naturally occurring speech data.