Adult learners of a second language often have difficulty with novel phonological categories, showing a bias towards existing (L1) category representations. In the production domain, this often results in a persistent accent, even when speakers are highly proficient in other aspects of the language. This study investigates the efficacy of a training paradigm that teaches the vocal tract configurations and articulatory targets for novel target sounds to new learners, with explicit reference to place, manner, and voicing features. Native English speakers learning Hindi coronal stop consonants (including a four-way voicing contrast, and a dental/retroflex place of articulation contrast) received visual cues and detailed descriptions of articulation targets, in addition to perceptual training. Participants' production of both place and voicing targets showed a significant improvement after production training (but not perception training) compared to baseline performance. However, only some voicing cues were maintained above baseline levels once cues were no longer present. These results suggest that this methodology has some promise, but limited retention and individual variation in performance indicate that some adjustment to the training may be warranted. I will discuss these issues, and some considerations for how they relate to both applied methods for language learning, and theoretical models of fledgling category representations.