Defense method and device for speech recognition model based on CycleGAN

A speech recognition model and model technology, applied in speech recognition, speech analysis, instruments, etc., can solve the problems of low recognition accuracy and resistance to sample attacks, and achieve the effect of improving recognition accuracy and resisting attacks

Active Publication Date: 2020-09-04
ZHEJIANG UNIV OF TECH
View PDF7 Cites 11 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0006] Aiming at the security problem that the current speech recognition system has low recognition accuracy and is vulnerable to attack by adversarial samples, the present invention provides a defense method for a speech recognition model based on CycleGAN, which can improve the recognition accuracy of the speech recognition model and can Enable the speech recognition model to resist the attack of adversarial samples, and improve the security and robustness of the speech recognition model

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Defense method and device for speech recognition model based on CycleGAN
  • Defense method and device for speech recognition model based on CycleGAN
  • Defense method and device for speech recognition model based on CycleGAN

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0032] In order to make the objectives, technical solutions and advantages of the present invention clearer, the following further describes the present invention in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, and do not limit the protection scope of the present invention.

[0033] The invention provides a defense method based on CycleGAN's speech recognition model, which specifically includes the following steps:

[0034] S1 generates a data set for training the CycleGAN model. The data set includes an adversarial sample data set and a normal speech data set, and the data set is divided into a training set and a test set.

[0035] The loss function of S11 initialized genetic algorithm is shown in formula (1):

[0036] L=l ctc (x, t) (1);

[0037] Where l ctc Indicates that the CTC loss function is used to measure how close the adversarial s...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses a defense method and device for a speech recognition model based on a CycleGAN. The defense method trains a CycleGAN model by utilizing an adversarial sample data set and a normal speech data set, GA-B generators in the model can perform denoising operation on the adversarial sample, and the normal sample is not influenced, and the normal sample is integrated into the speech recognition model to be used as a front-end processing device, so that the recognition precision of the speech recognition model is improved, the speech recognition model can resist the attack of the adversarial sample, and the robustness of the speech recognition model is improved.

Description

Technical field [0001] The invention belongs to the field of deep learning algorithms and information security research, and specifically relates to a defense method based on a CycleGAN-based voice recognition model. Background technique [0002] With the continuous development of technology, speech recognition technology is being used more and more by people. Speech recognition technology brings great convenience, simplifies the communication between humans and machines, omits the steps of intermediate keyboard input and handwriting, and has rich application scenarios. In the field of intelligent household appliances, there are voice-controlled household appliances, and people can control all voice devices in the home with voice; in the field of national defense, the automatic voice recognition system provides selected cockpit control through voice commands, which brings convenience to pilots ; In the medical field, voice recognition technology is used to help patients with sig...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G10L15/06G10L21/0208G10L15/20
CPCG10L15/063G10L21/0208G10L15/20
Inventor 陈晋音叶林辉
Owner ZHEJIANG UNIV OF TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap