End-to-end speaker verification method based on dual convolutional capsule network

CN114093372BInactive Publication Date: 2024-11-29SOUTH CHINA UNIV OF TECH
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202111195754.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-10-14
Publication Date
2024-11-29
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The prior art is difficult to effectively characterize the spatial relationship between characteristic elements in the confirmation of the speaker, and the multi-module error accumulation problem is serious, resulting in low accuracy in phrase speech samples or noise-containing situations.

Method used

The end-to-end speaker confirmation method based on the dual convolution capsule network is adopted to extract the features of different receptive fields through the feature extraction layer and the feature fusion layer, and the spatial information between feature elements is learned by using the dynamic routing layer to reduce multi-module errors. accumulation.

Benefits of technology

It effectively improves the accuracy of the speaker's confirmation, especially in the case of phrase speech samples or noise, reducing error accumulation and achieving higher recognition accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114093372B_ABST
    Figure CN114093372B_ABST
Patent Text Reader

Abstract

The present invention discloses an end-to-end speaker confirmation method based on a dual convolutional capsule network, the process is as follows: divide speech sample pairs; extract logarithmic Mel spectrum; construct a convolutional capsule network: including a feature extraction layer, a feature fusion layer, a dynamic routing layer and an L2 norm layer connected in sequence; train the dual convolutional capsule network: input the logarithmic Mel spectrum of the speech sample pairs in the training set into the convolutional capsule network to obtain two deep feature vectors, the deep feature splicing layer splices the two deep feature vectors and then inputs them into the decision output layer, iteratively updates the network parameters; speaker confirmation: input a test speech sample pair to determine whether the test speech sample pair is the same speaker. The present invention proposes a dual convolutional capsule network to make up for the defect that the convolutional network cannot effectively represent the spatial relationship of feature elements, and adopts an end-to-end confirmation method to directly obtain the speaker confirmation result from the input speech sample pair, eliminating the problem of inconsistent multi-module training targets and improving the accuracy of speaker confirmation.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Medicament module pharmacokinetic property and toxicity predicting method based on capsule network

    CN109979541A

  • Method for diagnosing planetary gearbox fault based on reinforced capsule network

    CN110657984A

  • Hyperspectral image classification method combining deep capsule network and Markov random field

    CN113128370A

  • KR20200019073A