Abstract: Systems and methods of the present invention may relate to prediction of a mental health score, representing a mental condition of a speaker. Embodiments of the invention may analyze speech by receiving an audio data element representing a discussion; extracting a first set of audio segments pertaining to speech of a first speaker in the discussion; analyzing the first set of audio segments to produce a set of audio features; and applying a machine-learning (ML) model on the set of audio features, to predict a mental health score, representing a mental condition of the first speaker. Embodiments of the invention may provide technical means for supporting psychiatric assessment of mental disorders such as depression and anxiety in an automatic, nondisruptive manner, e.g., without direct patient's collaboration, thereby increasing validity of the assessment.