This longitudinal study examined whether and to what extent the production of voice onset time (VOT) for /p/, /t/, and /k/ in English was improved in second language (L2) learners by enrolment in a content-based instruction (CBI) program at a Japanese university. Measures of VOTs were elicited at the beginning, middle, and end of the academic year and acoustically analyzed. Significant improvements in VOT production were observed over the year in which students received approximately 300 hours of input in the CBI setting. However, the results also showed that improvements in VOT were not affected by the amount of L2 input from native speakers. These findings suggest that L2 speech learning can occur if learners receive exposure to L2 in a CBI environment in the early stage of L2 learning through early, intensive exposure to L2, regardless of the source of input.