Emotional prosody is an important social cue that conveys the speaker's intention. Proper recognition of emotional prosody facilitates the interpretation of spoken language. Previous event-related potential (ERP) studies successfully documented an early negative (mismatch negativity, MMN) and a later positive (P3a) involuntary neural responses to detecting a change in emotional speech prosody (Schirmer et al., 2005; Zora et al., 2019). Nonetheless, these ERP components were elicited by controlling the linguistic content of the speech stimuli. It remains unclear whether natural affective prosody across varying linguistic carriers would elicit similar activation patterns. The current study adopted the multi-feature oddball paradigm to investigate the ERP responses to three basic emotional prosodies—happy, angry, and sad—embedded in varying monosyllabic English words. Thirty-three adult listeners (female = 23) completed the experiment. Unlike the previous reports, there was not a clear MMN response to the detection of emotional prosody patterns in the stimuli. But a late positive response (LPR) component was observed frontal-centrally in response to changes in affective prosody. Linear mixed-effect models further confirmed the presence of significantly larger LPRs to happy prosody than angry or sad prosody, suggesting that the LPR is a more reliable neural marker for emotional prosody recognition.