Speech Recognition of Isolated Arabic words via using Wavelet Transformation and Fuzzy Neural Network

Abstract

In this paper two new methods for feature extraction are presented for speech recognition the first method use a combination of  linear predictive coding technique(LPC) and skewness equation. The second one(WLPCC) use a combination of linear predictive coding technique(LPC),  discrete wavelet transform(DWT), and cpestrum analysis. The objective of this method is to enhance the performance of the proposed method by introducing more features from the signal. Neural Network(NN) and Neuro-Fuzzy Network are used in the proposed methods for classification. Test result show that the WLPCC method in the process of features extraction, and the neuro fuzzy network in the classification process had highest recognition rate for both the trained and non trained data. The proposed system has been built using MATLAB software and the data involve ten isolated Arabic words that are (الله، محمد، خديجة، ياسين، يتكلم، الشارقة، لندن، يسار، يمين، أحزان), for fifteen male speakers. The recognition rate of trained data is (97.8%) and non-trained data  is (81.1%).

Keywords: Speech Recognition, Feature Extraction, Linear Predictive Coding (LPC),Neural Network, Fuzzy network