A Comparative Study of Multilayer Feed-forward Neural Network and Radial Basis Function Neural Network Models for Speech Recognition

The most common way of human-to-human communication is speech. As speech provides the easiest and most natural way of interaction, it becomes the need of human-to-machine communication as well. Automatic speech recognition (ASR) is the technology to enable machines to understand process and recognize speech. Due to its applicability in various application domains, ASR becomes one of the most fascinating areas of pattern recognition. In this paper, we are analyzing the performances of multilayer feed-forward neural network and Radial basis function neural network models for the recognition of speech signals. The work is conducted in four stages: speech signal acquisition & pre-processing, feature pattern vector creation, implementation & training of selected neural network models and comparative analysis of the performances of selected neural networks. Proposed work is conducted with 10 speech samples of English alphabets .Digital signal processing operations are applied on signals to convert them and make them appropriate for further processing. Five feature pattern vectors are created to be used for training and testing of the network models. Performance of selected neutral network models is measured and analyzed for the created feature pattern vectors. Results indicate that feed-forward neural network model performs better than the Radial basis function neural network for all the test pattern vectors.