| Authors |
Berdibaeva Gul'mira Kuanyshbaevna, doctoral student, Kazakh National Technical University named after K.I.Satpayev (22 Satpayev street, Almaty, Kazakhstan), horli@mail.ru
Bodin Oleg Nikolaevich, doctor of technical sciences, professor, sub-department of information and measuring equipment and metrology, Penza State University (40 Krasnaya street, Penza, Russia), bodin_o@inbox.ru
Gromkov Nikolay Valentinovich, doctor of technical sciences, professor, sub-department of information and measuring equipment and metrology, Penza State University (40 Krasnaya street, Penza, Russia), ngrom@bk.ru
Kozlov Valeriy Valer'evich, candidate of technical sciences, associate professor, sub-department of information and measuring equipment and metrology, Penza State University (40 Krasnaya street, Penza, Russia), iit@pnzgu.ru
Ozhikenov Kasymbek Adilbekovich, candidate of technical sciences, associate professor, head of sub-department of robotics and automation equipment, Kazakh National Technical University named after K.ISatpayev (22 Satpayev street, Almaty, Kazakhstan), horli@mail.ru
Pizhonkov Yaroslav Andreevich, student, Penza State University (40 Krasnaya street, Penza, Russia), iit@pnzgu.ru
|
| Abstract |
Background. It is considered a possibility application of artificial neural networks for voice recognition in voice control applications.
Materials and methods. The applied method is based on the analysis of the voice commands, neural networks, each trained on respective phonemes of a natural language. Selected segments of speech commands in parallel applied to the input of each network. In the case of the recognition of phonemes corresponding to the network emit a signal, then a group of recognized phonemes are analyzed and compared with the basic vocabulary of voice commands.
Results. The article considers modern methods and means of voice control, the analysis of which showed the need to use phonemic decoding method in the speech recognition. It is provided, an example, of the work of separation of a voice message into its phonemes. By converting a voice command is divided into phonemes, which are the input to the neural network analysis.
Conclusions. The proposed approach in the construction of systems increases the accuracy of voice recognition by identifying the frequency characteristics of phonemes and their subsequent neural network analysis. This increases the susceptibility of the neural network to the input data in comparison with the direct input of the neural network speech commands.
|
