Human Voice Recognition by Using Hebb Artificial Neural Network Method

Abstract

It has been created a program to recognize human voice by using artificial neural network (ANN). The ANN method used is Hebb. Hebb was chosen because it is the simplest ANN so the training and testing process is faster than other methods. Designing the program is started by designing Hebb’s architecture and design of GUI (Graphical User Interface) using Matlab R2009a. The design of Hebb's architecture consists of 4500 inputs and 3 outputs. The GUI design of the program consists of three main sections: recording panels to record sample sounds, training panels to determine the weighted value and bias of the training results according to the Hebb training algorithm, and the testing panel to test the test sounds according to the Hebb testing algorithm. After program design, proceed with the testing of the program. Testing of the program starts with the sound recording of samples from 8 different people using the record panel. Each person has 1 voice sample for training data. Then proceed with the Hebb training process using the training panel, weight and bias value displayed on the training panel. After the weight and bias values ??are obtained, proceed with the Hebb testing process using 16 test sound data consisting of 8 sound data equal to training data and 8 noise data. From the testing program process obtained a result of 100% for the level of recognition of the same voice data with training data and for noise data has a recognition rate of 87.5%.