This study assesses the efficacy of short, unsupervised speech training with the mobile app SayEst. The data from production, perception, and self-assessment questionnaires of 30 Russian L1 speakers from different proficiency levels were collected before and after using the app. Although there were no significant differences between the pre- and post-test conditions, the results showed that learners' confidence in speaking in Estonian rose after the training. The beginner learners, in general, made more mistakes than the intermediate and advanced learners. The learners’ mistakes in perception and production reflected the differences and similarities in the phoneme inventories between Estonian and Russian. The pre- and post-test recordings were analyzed by four expert phoneticians who rated the accentedness of each of the phonemes in words on a scale from 1-3. The participants’ recordings and ratings of the annotators will be used to train an L2 speech verifier system that will give the user feedback on each of the segments in a word. The annotators noted that the same vowels that were perceptually difficult for the learners were also difficult to produce. With consonants, the annotators noted problems with plosives, /h/, /l/ and /j/. SayEst helps to raise learners’ awareness of their pronunciation and build up their confidence. SayEst is recommended as supplementary material as unsupervised training did not have the hoped effect in perception training. In future studies, we will focus on production and describe the effects of unsupervised training.