Speech recognition based on convolutional neural networks

Speech recognition, as the man-machine interface, plays a very important role in the field of artificial intelligence. Traditional speech recognition methods are shallow learning structure, and have their limitations. This paper uses the Convolution Neural Networks (CNNs) to realize speech recognition. It is an alternative type of neural network that can reduce spectral variation and model spectral correlations which exist in signals. Besides the paper uses Back Propagation to train the neural network. During the whole experiment, the paper uses a group of speech that recorded by ourselves as training data, and it uses the others to test the neural network. Experimental results show that CNNs can efficiently implement isolated word recognition.