With recent technical advances in robotics and artificial intelligence, humanoids have proven to be valuable tools to investigate mechanisms of human social interactions (Wykowska et al., 2016). While many behavioural features are relevant for social interactions, voices, together with faces, are among the most impactful (Belin, 2018) but also the less studied (Fischer & Niebuhr, 2023). While their responsiveness of temporal vocal areas (TVAs) to vocal stimulations is now clearly established, little is known about the factors that impact their response (Belin et al., 2000). It suggests that response in Temporal Voice Areas increases as individual voices diverges from an average human voice (Latinus et al., 2013). Here, we sought to investigate the response in Temporal Voice Areas to synthetic voices. As these voices are significantly different from human voices, norm-based coding suggests an increased response in TVAs. We performed an analysis using an existing corpus of 25 participants recorded with functional MRI while engaging in conversations with human and robot interlocutors, the later endowed with a text-to-speech synthesis. We focused on periods when participants listened to the interlocutor. Results (Figure, left) indicate that TVAs identified by an increased response to human voices compared to the implicit baseline show an additional increased response to synthetic compared to human voices. Reciprocal comparisons between the artificial and natural interlocutors (Figure, right) revealed that the temporoparietal junctions associated with mental states attributions is more active when listening to the former. Altogether, our results suggest Temporal Voice Areas increased activity signals synthetic voices as unnatural.