In this study, we conduct a longitudinal experiment to explore the effects of adaptive phonetic training on the identification of the English vowels /i/ and /I/ by learners in China. The vowel /i/ in English has a low F1 value, a high F2 value, and a long duration. In contrast, the vowel /i/ has a high F1 value, a low F2 value, and a short duration. Our experiment was designed to determine whether and how learners can establish a new category for the English vowel /I/. A total of 30 participants were trained on computers under the guidance of instructors for 6 months. The correctness ratio of their perception of /i/ and /I/ is examined in the pre-tests, post-tests, and after-tests. The results support Escudero's Hypothesis. Learning a new phoneme includes four stages: inability to distinguish, duration-based, duration and spectral-based, and L1-English-like, primarily spectral-based. We use our findings to develop an improved hypothesis regarding the range of duration of training for learners. We found that phonetic training with instructions helped second/foreign language learners establish a new category for the English vowel /I/, with the same results as those for learners acquiring the language in English-speaking environments, but with different boundaries of categories from those of native speakers. This phenomenon may align with the dissimilation in the phonetic category that has been reported for early bilingual learners in past research. The categorical boundary used by second/foreign language learners along the spectral dimension is farther from the center of the category of a vowel similar to the vowel being learned that exists in their first language.