Deception voice detection method based on deep neural network

A technology of deep neural network and voice detection, which is applied in the field of deceptive voice detection based on deep neural network, can solve problems such as ignoring spoofing attacks and using deep neural network, and achieve enhanced generalization ability, reduced interference, and strong nonlinear construction. The effect of modeling ability

Active Publication Date: 2019-11-22
XIAMEN UNIV
View PDF15 Cites 31 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

This invention only detects voice playback, and does not consider complex spoofing attacks and the use of deep neural networks

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
  • Deception voice detection method based on deep neural network
  • Deception voice detection method based on deep neural network
  • Deception voice detection method based on deep neural network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0057] Please check figure 1 and figure 2 , a flow chart of a method for deceiving voice detection based on a deep neural network, comprising:

[0058] Set up the fraudulent speech detection model step, set up the fraudulent speech detection model based on deep neural network according to the existing known true and false speech data of the user;

[0059] For the judgment step, choose "step 203" or "step 204 to step 208".

[0060] The described steps of setting up a fraudulent speech detection model include:

[0061] Step 201, extracting the acoustic features of all speech samples in the training set, the acoustic features include MFCC features;

[0062] Step 202, the acoustic feature that will extract from training set is sent into neural network, trains the network parameter of whole neural network according to task cost function, after neural network training is finished, fixed network parameter, namely, has set up fraudulent speech detection model;

[0063] Step 203, ...

Embodiment 2

[0073] Please check image 3 , a flow chart of a method for deceiving voice detection based on a deep neural network, comprising:

[0074] Step 301, acoustic feature extraction and training sample generation step, which includes:

[0075] First extract at least two acoustic features of high time-frequency resolution of all speech samples in the training set, such as MFCC and FBank, that is, for Mel Frequency Cepstral Coefficient (Mel Frequency Cepstral Coefficient, MFCC) and Mel (Mel) frequency filter bank (Mel Frequency Bank, FBank). Set the corresponding extraction frame shift from the general 10 milliseconds to 4 milliseconds, the number of filter banks per frame is changed from the general 30 to 160, and the discrete cosine transform is changed from the general 20 dimensions to 40 dimensions with high time-frequency resolution rate of MFCC acoustic characteristics;

[0076] After each frame is aligned, the MFCC and FBank acoustic feature vectors of each frame are sequen...

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 deception voice detection method based on a deep neural network. The method comprises the following steps: A, training and building a deception voice detection model based onthe deep neural network according to voice data, with the known authenticity, of a user, wherein the deception voice detection model has network parameters; and B, performing classification judgmenton test voices to be tested in the trained deception voice detection model with the network parameters so as to judge whether the test voices are real voices or deceptive voices. The method has the following advantages: novel unknown voice synthesis, voice conversion, record playback and other deceptive attacks are supported.

Description

technical field [0001] The invention relates to the technical field of computer information services, in particular to a method for detecting fraudulent voice based on a deep neural network. Background technique [0002] Speaker recognition is to identify a person's identity from the speaker's voice. In layman's terms, it is answering the question "Who is talking?" Specifically, the distinguishable voiceprint representation of the individual is extracted from the speaker's voice, and the representation is used as the speaker's identity information to achieve identification. In practical application scenarios, speaker recognition technology, like other identity verification technologies, is accompanied by artificial malicious deception attacks, which has security problems. [0003] Currently, there are three main spoofing attack modes: [0004] (1) Deliberate imitation from other speakers (such as ventriloquist skills); [0005] (2) Natural speech synthesized by high-qual...

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): G10L17/02G10L17/04G10L17/14G10L17/18G10L17/22
CPCG10L17/02G10L17/04G10L17/14G10L17/18G10L17/22
Inventor 李琳黎荣晋洪青阳
Owner XIAMEN UNIV
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