Models of speech learning suggest that adaptations to foreign language sound categories should happen early in the acquisition process. Results from laboratory language training show effects on non-native perception within one to three weeks of training. Results from linguistic immersion studies show differences in adaptations when contrasting averages of 1-2 yrs of experience with 6-7 yrs of experience. We investigated this apparent discrepancy in a longitudinal study on Danish language officer cadets learning either Arabic (MSA and Egyptian dialect) or Dari (Afghan Farsi) through intensive multi-faceted language training. We conducted two experiments (identification and discrimination) with the two groups of cadets. The cadets were tested at four time points (at the start (T0), after three weeks (T1), after six months (T2), and after 20 months (T3)). We used a phonemic Arabic contrast (emphatic frication) and a phonemic Dari contrast (fricative voicing) as stimuli for both groups. We saw an effect of learning on the Dari learners’ identification of the Dari stimuli already after three weeks of language training, which was sustained, but not improved, after six and 20 months. The extents of the changes in the Dari learners’ identification functions were positively correlated with their grades from a spoken exam after six months. No other learning effects at the group level were observed, i.e. no effects of learning on the participants’ categorical perception within the long-term training (six to 20 months). We thus show adaptations to a foreign voicing contrast within just three weeks of intensive foreign language learning, but no further improvement after six and 20 months.