<p>Emotions represent the mental state of a person. It plays a vital role in our day-to-day life. As technology is evolving rapidly, the scope and applications of emotion recognition from the speech are also increasing by leaps and bounds. A lot of progress has been made in this area but still, there is ample scope for improvement because to recognize emotions, the machine has to overcome a number of challenging aspects such as linguistic barriers, accuracy, and speed of recognition in real-time. In addition, emotion is subjective and varies from individual to individual. In this paper, we have proposed a model using inception version-4 network for deep features extraction and ensemble classifier namely Random forest (RF) classifier for classification of speech. The emotions are classified to Plutchik’s eight basic emotions. We tested our model on the RADVESS dataset having 8 emotions and the SAVEE dataset having 7 emotions. We have achieved an accuracy of 97.57% and 90.62% respectively. The proposed model may serve as a baseline for propelling further research in the direction of a new form of feature extraction and classification of emotions.</p>