Feature extraction is an important part in speech emotion recognition, and in relevance feature extraction in speech emotion recognition problems, this paper projected a fresh technique of feature extraction, victimization DBNs in DNN to extract emotional options in speech signal automatically. By implementing a 9 layers depth DBNs, to extract speech emotion feature and incorporate multiple consecutive frames to create a high dimensional feature. An improved CNN model is presented here which consists of combination of convolution 1d layers and generalized to form a 9 layer architecture of CNN (convolutional neural network), model accuracy has been checked with respect to emotion classes such as considering 5 emotions considered as angry, calm, fearful, happy, sad for male as well as female and eventually speech emotion recognition multiple classifier system was achieved. The speech feeling recognition rate of the system reached 89.00% that was approx. 14% additional over the conventional technique.