The experiment reported here investigated the neural correlates of predictive processing of angry and neutral speech. Twenty-six participants listened to recordings of angry and neutral conversation segments, as well as to speech-shaped noise, while their EEG was recorded. Oscillatory power in the gamma band (30–80 Hz) and the N400 component of event-related potentials (ERP) to sentence-final words were analyzed. In comparison to neutral words, negative emotional valence significantly reduced the amplitude of the N400 elicited by sentence-final words. Furthermore, there was larger gamma power during exposure to angry speech in comparison to neutral speech. The results generally suggest increased prediction and facilitated semantic integration in negative as compared to neutral speech. To date, the predictability effects on gamma power have been reported with relation to the semantic-lexical content of words. The present findings demonstrate that gamma power is also modulated by the emotional content of speech.