This dissertation reports on perceptual training of tense-lax vowel contrasts in the context of an advanced-level ESL pronunciation class for Chinese and Korean international graduate students. The vowel contrasts were trained under four training paradigms designed to examine the effects of variation due to multiple speakers, different speech rates and coda consonant. Training material consisted of nonce word minimal pairs used to mitigate task complexity related to lexical access and to circumvent the effects of frequency and top-down processing. Participants completed pre- and post- tests on discriminating vowels in real word minimal pairs and nonce word minimal pairs. Vowel perceptual training took place over the course of six days wherein each day’s training consisted of 100 exemplars that students played as much or as little as they wanted; the number of sound files they played was tracked and recorded. Participants finished training at their own discretion and then were tested on a 25-member subset of that day’s training tokens. Chinese and Korean learners trained under one of four training paradigms; Training Paradigm A, multiple speakers, three speech rates; B, multiple speakers, one speech rate; C one