Neuro Based Approach for Speech Recognition by Using Mel-frequency Cepstral Coefficients

NEURO BASED APPROACH FOR SPEECH RECOGNITION BY USING MEL-FREQUENCY CEPSTRAL COEFFICIENTS R.L.K. Venkateswarlu1 and R. Vasanthakumari2 1 Department of Information Technology, Sasi Institute of Technology and Engineering, Tadepalligudem, India, E-mail: [email protected]. 2 Perunthalaivar Kamarajar Arts College, Puducherry-605107, India, E-mail: [email protected]. This paper presents continuous speech recognition system based on neural network concept. Features are extracted and the data is compressed using Mel-frequency Cepstral coefficients method. These Mel-frequency Cepstral coefficients are used as inputs to train neural networks. Neural networks are useful to solve complex problems which do not require accurate solution. The backpropagation algorithm is used in multilayer perceptron. The solution found in this approach is convergent. This research work is aimed at speech recognition using multilayer perceptron neural networks. A small vocabulary of 11 words were established first, these words are upload, search, browse, import, export, send, remove, attach, help, format, install. These chosen words involved with executing some computer functions such as export a file or an image; find a file or a folder or a image; to view the data; download some properties; transfer out of a database or document in a format; to move a file; to delete a file; to add a file; some assistance; to delete existing content; to add new software. Are introduced to the computer and then subjected to feature extraction process using Mel-frequency cepstral coefficients. These features are used as input to an artificial neural network in speaker dependent mode. Half of the words are used for training the artificial neural network and the other half are used for testing the system. The system components consist of three parts, speech processing, feature extraction, training and testing by using neural networks and information retrieval. The retrieve process proved to be 81.44%–93.18% successful, which is quite acceptable, considering the variation to surroundings, state of the person, and the microphone type.