Human speech contains different emotions such as joy, sorrow, anger, surprise, calm, disgust, and fear. These emotions may vary according to the situation. So, emotions in human speech play an important role in expressing feelings and communicating with others. One should try to understand the emotions and respect one’s emotions. Emotion is the cornerstone in human–computer and human–robot interactions. The main difference between humans and humanoids is emotion. Humans can express their emotions through speech, actions, and facial expressions. Different people have different emotions and altogether different ways to express it. Therefore, emotion declaration is a challenging task in the vision of computers. This chapter proposes a convolutional recurrent neural network (CRNN) for emotion recognition in speech. CRNN is the hybrid model, which combines 3D CNN and GRU. A 3D convolutional gated recurrent unit (CGRU) is proposed to recognize and classify emotions from human speech. This neural network extracts spatial features and temporal dependencies from human speech and classifies them into one predefined emotion. Human speech is converted into spectrograms and Mel-frequency cepstral coefficients (MFCCs). CNN will learn spatial patterns in spectrograms by using convolutional filters and then the output is given to GRU to capture the temporal dependencies. This hybrid model allows to leveraging of both types of patterns for accurate emotion recognition. Our experimental results on the RAVDESS dataset reveal that our suggested model achieved an accuracy of 97%.