A Study on the Differences in Vowel Duration in Japanese in Natural and Written Speech: An Empirical Analysis Based on a Corpus

Abstract

Japanese vowel duration is a key phonetic feature for distinguishing lexical meaning and stylistic register, but its realization patterns across different speech contexts remain insufficiently quantified. This study constructs a bimodal corpus involving 20 native Japanese speakers, including 50 minutes of spontaneous conversation and 30 minutes of news text reading. Using Praat software, duration data for the five Japanese vowels /a, i, u, e, o/ are annotated, producing 8,347 valid samples. The analysis first extracts absolute vowel duration within syllable boundaries, then calculates the long-to-short ratio, applies mixed-effects modeling to control individual speaker differences, and uses independent-sample t-tests to verify inter-group significance. The results show that word-final vowels in conversational contexts are shortened by 28.4% compared with written-language reading, with /u/ showing the most significant shortening rate of 31.7%. Prosodic boundary location and speech-rate variation are the main factors affecting duration differences. The findings provide acoustic signal evidence for natural speech training, Japanese prosody instruction, and prosodic modeling in speech synthesis systems.