Emotions are transferred through both verbal and non-verbal indications. But visually impaired people are unable to perceive non-verbal cues of conversation patterns. To detect emotions automatically, it is necessary to capture these verbal and non-verbal indications. Hence, researchers are concentrating on enhancing the performance of multimodal emotion recognition(MER), where various modalities like text, speech, face, body gesture, physiological signals, etc. are taken into consideration as an input simultaneously. This review explores a general framework of the MER system in brief. It also deliberates on the significance of feature extraction in the MER systems along with the types and importance of multimodal fusion methodologies. Also, we have discussed the current state of MER research. Various popular and open-source datasets used to build MER systems are also described in this review.