METHOD OF NEURAL NETWORK ANALYSIS OF VOICE SIGNAL

Abstract

The article is devoted to increase of efficiency of the means of analysis of biometric characteristics of subjects, interacting with information-control systems of various purpose. It is shown that from the standpoint of the possibility of using as a sensor the reading of the biometric parameters of the standard peripheral equipment of the computer, the widespread use in the information-control systems of voice messages, their high informativeness, the complexity of falsification of biometric information, as well as the possibility of carrying out hidden monitoring, the prospects have analysis tools voice signal. The necessity of improvement of methodology of neural network analysis of voice signal for recognition of emotions and person is grounded. Possibility of increase of efficiency of neural network means of analysis due to adaptation of parameters of neural network model to the conditions of use is determined. The principles of determination of valid neural network models and the most effective type of neural network model used for voice signal analysis have been formed. A coding procedure for the input signal is developed, which allows to use in the neural network a compact representation of the most informative features of a voice signal. A procedure for encoding a neural network output signal has also been developed to increase the efficiency of its learning. The method of neural network analysis of the voice signal is developed, which due to the offered principles of adaptation and procedures of coding of input and output parameters, allows to introduce into neural means a neural network whose architecture is adapted to the expected conditions of use. The effectiveness of the proposed method has been proven experimentally. Experimental studies have shown that the use of the developed method allows to ensure the accuracy of recognition of emotions of the identified speaker, which is approximately 0.94, which corresponds to the best modern decisions in this field. It is suggested to correlate the ways of further research with the development of solutions that would allow to analyze the voice signal of arbitrary duration under the conditions of noise of different kind.