To obtain access to full text of journal and articles you must register!
- Article name
- Determination of the parameters of a training sample for a neural network used to determine the percentage of speech intelligibility in tasks of assessing the security of speech acoustic information
- Authors
- Volkov N. A., , volkovnikandr@gmail.com, Samara State Technical University, Samara, Russia
Ivanov A. V., , andrej.ivanov@corp.nstu.ru, Samara State Technical University, Samara, Russia
- Keywords
- deep neural networks / convolutional neural networks / signal-to-noise ratio / audio recording noise / speech recognition / spectrograms / low-frequency cepstral coefficients / assessment of the security of speech acoustic information
- Year
- 2024 Issue 4 Pages 44 - 51
- Code EDN
- XZYMKN
- Code DOI
- 10.52190/2073-2600_2024_4_44
- Abstract
- To solve the problems of assessing the security of speech acoustic information, it is proposed to use convolutional neural networks. The work is devoted to the study of the most suitable parameters of spectrograms and low-frequency cepstral coefficients generated on the basis of audio recordings of speech with superimposed white noise to create a training sample used in training a convolutional neural network. The parameters of the convolutional neural network model are determined, and the requirements for the data set for its training are formed. In this article, the authors change one of the parameters in the training sample in order to determine the most appropriate values. As a result of the study, it was concluded that the format of representing data sets in the form of graphs of low-frequency cepstral coefficients is best suited for solving such a problem. In the future, it is planned to expand the data set by increasing the number of speakers and adding various masking noise spectrum.
- Text
- BUY for read the full text of article
- Buy
- 500.00 rub
