Japanese EFL learners frequently experience difficulty producing American English (AE) vowels, which are perceptually assimilated into native Japanese categories. While previous research has shown that auditory-based training can improve vowel identification, few studies have examined whether multimodal CALL-based training leads to measurable changes in learners’ vowel production. This study extends our earlier work using JSpace, an interactive web-based vowel-space training tool that integrates auditory input with visual representations of vowel quality and explores its role in supporting L2 phonological restructuring through multimodal input. Sixty Japanese university students were assigned to either a multimodal JSpace training group or a control group receiving traditional auditory-only identification training during a four-week period. Both groups completed identical pretest and posttest discrimination and production tasks targeting seven AE vowels presented in CVC contexts. In addition to perceptual measures, this study focuses on an acoustic analysis of learners’ vowel productions. Changes in first and second formant (F1–F2) frequencies were examined to assess pre- to post-training shifts in the acoustic vowel space. Results show that learners trained with JSpace exhibited greater shifts toward native AE vowel targets than the control group, indicating systematic restructuring of L2 vowel production. Pedagogical implications for CALL-based pronunciation instruction are discussed.