Search academic texts

Information × Registration Number 0221U106169, 0121U110859 , R & D reports Title Methods and means of recognizing the emotions of the speaker popup.stage_title Head Tereikovska Liudmyla О., Registration Date 29-10-2021 Organization Kyiv National University of Construction and Architecture popup.description2  As a result of the analysis of literary sources, the necessity of improving the methodology of neural network analysis of the voice signal for the recognition of emotions has been substantiated. The possibility of increasing the efficiency of neural network analysis tools by adapting the parameters of the neural network model to the conditions of use has been determined. The principles of defining admissible neural network models and the most effective type of neural network model have been formed; A procedure for encoding an input signal has been developed, which ensures the use of a compact representation of the most informative features of a speech signal in a neural network. Also, a procedure for coding the output signal of a neural network has been developed, which makes it possible to increase the efficiency of its training. A method of neural network analysis of a voice signal has been developed, due to the proposed principles of adaptation and coding procedures for input and output parameters, it allows to implement a neural network into recognition tools, the architecture of which 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 makes it possible to ensure the recognition accuracy of the identified speaker's emotions, equal to about 0.94, which corresponds to the best modern solution in this area. It is proposed to correlate the ways of further research with the development of solutions that would allow analyzing a voice signal of arbitrary duration under noise conditions. Product Description popup.authors Solovei Olga Leonidivna popup.nrat_date 2021-10-29 Close