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

Unsupervised active learning method for speech emotion calculation

A technology of emotional computing and active learning, applied in the field of emotional computing, to achieve the effect of wide application range, convenient use, and few restrictions

Active Publication Date: 2020-02-11
HUAZHONG UNIV OF SCI & TECH
View PDF3 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problem that the existing supervised active learning method for speech emotion computing needs to know a small number of real labels and needs to interact with human experts multiple times, the present invention provides an unsupervised active learning method for speech emotion computing, Its purpose is to release the existing supervised active learning regression algorithm for speech emotion computing, which needs to know a small number of real labels and need to interact with human experts multiple times.

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
  • Unsupervised active learning method for speech emotion calculation
  • Unsupervised active learning method for speech emotion calculation

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1—— 1

[0049] Embodiment 1—one-step optimization

[0050] Such as figure 1 As shown, the method includes the following steps:

[0051] Step S1. Obtain the dimension d=46 of the feature vector of the samples in the VAM data set, set the number of voice samples handed over to experts for marking as M=1, and set the maximum value of the number of iterations c as c max =5, the optimization method is set to the linear manifold method (see later description for details).

[0052] This embodiment preferably c max =5.

[0053] Step S2. Use Principal Component Analysis (PCA) to reduce the dimension of the VAM data set to d=M−1=9 dimensions.

[0054] This is to perform feature processing on the VAM data set so as to implement the linear manifold method. The steps after Embodiment 1 are all calculated based on the VAM data set after dimensionality reduction. m 1 =M=d+1=10.

[0055] Step S3. The number of iterations c is initialized to 1, and the sample set to be marked corresponding to t...

Embodiment 2

[0064] Example 2 - two-step optimization.

[0065] Such as figure 2 As shown, the method includes the following steps:

[0066] Step S1. Obtain the dimension d=46 of the feature vector of the sample in the VAM data set, set the number of speech samples handed over to experts for marking as M=60, and set the maximum value of the number of iterations c as c max = 5, the optimization method of the first step is set to the linear manifold method, the optimization method of the second step is set to the greedy search method (see later description for details), and the first step selects M 1 =d+1=47 samples to be marked, choose M in the second step 2 =M-M 1 = 13 samples.

[0067] The sum of the number of samples to be marked selected in each step must be equal to M, that is, M 1 +M 2 =M.

[0068] Step S2. Start to execute the first step optimization of the algorithm. The number of iterations c is initialized to 1, and the sample set to be marked corresponding to the cth ite...

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 an unsupervised active learning method for speech emotion calculation, and belongs to the field of emotion calculation. Compared with an existing supervised active learning method for speech emotion calculation, in which a small number of known real tags are needed, and repeated interaction with an artificial expert is needed; the invention method performs multi-step iterative optimization, the samples in a to-be-tagged sample set Sc are optimized in sequence in each round of iteration; only one sample is optimized each time; each sample in each round of iteration is only optimized once; a small number of most valuable to-be-tagged samples are selected; the samples can better represent all the samples in the sample set, and no known tag or existing regression modelis needed; the method is suitable for a voice data set without tags completely, repeated interaction with experts is not needed, all to-be-tagged voice samples can be provided for the experts at a time, the limiting conditions are fewer, the application range is wider, and use is more convenient.

Description

technical field [0001] The invention belongs to the field of emotional computing, and more specifically relates to an unsupervised active learning method for speech emotional computing. Background technique [0002] Emotional computing is an artificial intelligence technology that automatically recognizes human emotions through computers. Speech emotional computing refers to inputting people's words into the computer, and then extracting the features of the original speech signal by the computer, inputting it into the machine learning model to obtain the predicted output, and then Get the emotion in this sentence through emotion decoding. Speech emotion coding generally uses continuous values ​​to encode the degree of emotion (for example, 0-1 means calm to very angry), so the machine learning model used is a regression model. This technology enables computers to understand the emotions hidden in human speech, thereby expanding the functions of human-computer interaction. ...

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
IPC IPC(8): G10L15/02G10L15/08G10L25/27G10L25/63
CPCG10L15/02G10L15/08G10L25/27G10L25/63
Inventor 伍冬睿刘子昂
Owner HUAZHONG UNIV OF SCI & TECH