The accuracy of voice or speech recognition is affected due to the presence of various background noises present in the surroundings. Automatic Speech Recognition communication systems are utilized for enhancing the speech by either reducing or eliminating the surrounding noises. The corrupted speech signals are enhanced by using different techniques. In this paper, Recurrent Convolutional Encoder-Decoder (R-CED) network is proposed for enhancing the speech by the elimination of noise signals. The efficiency of the proposed work is determined by comparing the performance metrics like PESQ, STOI and CER with various existing techniques. From the results obtained, it can be confirmed that the efficiency of proposed R-CED is higher and optimal when compared to the existing techniques.