Skip to Main Content
Neural networks have found profound success in the area of pattern recognition. In the recent years there has been use of neural network for speech recognition. In this paper backpropagation neural network has been used for isolated spoken Urdu digits recognition. Mel frequency cepstral coefficients (MFCC) has been used to represent speech signal. Dimensions of speech features were reduced to a vector of 39 values. Only 39 values from MFCC features speech are fed to the neural network having more than one hidden layers with varying number of neurons, for training and recognition An analysis has been made between different number of hidden layers and different number of neurons on hidden layers. It has been found that results for these 39 values are similar to that obtained using complete MFCC features that range from 804 to 67x39. With the use of 39 values on input layer, computational complexity and time for training and recognition of neural network is reduced. In order to evaluate the significance of the proposed method on data other than Urdu digits, 30 English words have been trained and recognized that gave 98% results. All the implementation has been done inMATLAB.
Date of Conference: 14-17 Aug. 2007