Accurate speech-sound perception and production are prerequisites for communication in a non-native language. Earlier research has shown that new categorization and pronunciation patterns evolve in different learning settings and that these skills can be trained with various methods. We tested the effects of two types of training protocols on the production and identification of L2 vowels embedded in a pseudo-word context. Group 1 (Producers) participated in a listen and repeat training, where they produced the target stimulus /tᵾ:ti/ and the non-target stimulus /ty:ti/ after the example in a pseudo-randomized order; Group 2 (Listeners) was instructed to count the number of targets /tᵾ:ti/ in the same stimulus train without any motoric production movements. The results showed clearly that listen and repeat training led to plastic changes both in production and in identification, while no learning effects were obtained with the listening paradigm. This suggests a significant role of motoric experience in the acquisition of speech.