Unlock instant, AI-driven research and patent intelligence for your innovation.

A Speaker Verification Method Based on Deep Mixture Model

A technology of speaker confirmation and mixed model, which is applied in speech analysis, instruments, etc., can solve problems such as not considering the nature of data derivatives, and achieve the effect of improving accuracy

Active Publication Date: 2021-10-01
TSINGHUA UNIV
View PDF4 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to overcome the shortcoming that the traditional Gaussian mixture model does not consider the derivative nature of the data when modeling the speaker, and propose a speaker confirmation method based on the deep mixture 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
  • A Speaker Verification Method Based on Deep Mixture Model
  • A Speaker Verification Method Based on Deep Mixture Model
  • A Speaker Verification Method Based on Deep Mixture Model

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0044] The present invention proposes a speaker confirmation method based on a deep hybrid model, which will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0045] The present invention proposes a speaker confirmation method based on a deep hybrid model, comprising the following steps:

[0046] 1) Establish a deep mixture model of the speaker, the process is as follows figure 1 As shown, the specific steps are as follows:

[0047] 1.1) Obtain the training speech data of the speaker;

[0048] In this embodiment, the speech data of a target speaker in the NISTsre16 evaluation is used as the training speech data. The number of training voice data is 1-5 pieces, and the length of each piece is 10 seconds-120 seconds. Each piece of training data is known to be the voice of the target speaker.

[0049] 1.2) Preprocessing the training speech data, extracting step 1.1) M D-dimensional Mel cepstrum feature sets correspon...

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 proposes a speaker confirmation method based on a deep mixed model, which belongs to the technical fields of voiceprint recognition, pattern recognition and machine learning. This method first obtains the training speech data of the target speaker and performs preprocessing to obtain the Mel cepstrum feature set of the training speech data; establishes the 0th layer Gaussian mixture model for the Mel cepstrum feature set and obtains the first derivative set; then establish the first layer of Gaussian mixture model and the second layer of Gaussian mixture model in turn, until the establishment of the S-th layer of Gaussian mixture model, combine all Gaussian mixture models to obtain the depth mixture model of the speaker; then obtain the test voice data and extract Mel cepstrum feature set to establish a deep mixture model of the test speech data; calculate the similarity between the two models: if the similarity is less than or equal to the threshold, the test speech data belongs to the target speaker. The present invention not only considers the distribution of the data itself, but also considers the distribution of the derivative of the data, so as to improve the accuracy of speaker confirmation.

Description

technical field [0001] The invention belongs to the technical fields of voiceprint recognition, pattern recognition and machine learning, and in particular relates to a speaker confirmation method based on a deep mixture model (DMM). Background technique [0002] Speaker confirmation refers to judging whether a test voice is spoken by a designated speaker. With the rapid development of information technology and communication technology, speaker verification technology has been paid more and more attention and widely used in many fields. Such as identification, apprehension of criminals in the telephone channel, identity confirmation in court based on telephone recordings, telephone voice tracking, and anti-theft door opening functions. In the field of Internet applications and communications, speaker confirmation technology can be applied to voice dialing, telephone banking, telephone shopping, database access, information services, voice e-mail, security control, computer...

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 Patents(China)
IPC IPC(8): G10L17/04G10L17/02G10L25/24
CPCG10L17/02G10L17/04G10L25/24
Inventor 何亮陈仙红徐灿梁天宇刘加
Owner TSINGHUA UNIV