Learning outcomes with text-to-speech synthesis of native and non-native English varieties

English is a common instruction medium of learning tools using Text-to-Speech (TTS), yet most solutions incorporate only L1 varieties like Standard American English (AmE). At the same time, some research suggests that educational content personalized in the learner's variety is beneficial. We tested this hypothesis with students from the Masaryk University Brno, who listened to a lecture synthesized in Czech English (CzE) and AmE, rated the TTS speaker based on the Robotic Social Attributes Scale and answered questions regarding the contents of the lecture. The learners improved their knowledge similarly with both TTS varieties. Characteristics from the intelligence cluster were rated higher than anthropomorphism and likability, and the AmE voice was rated more competent than the CzE voice. While the results indicate that the narration may need to be made more engaging, the present study provides first insights into the perceptions of L2 students' own variety and recommendations for customizing credible, learning-facilitating TTS.