Eureka AIR delivers breakthrough ideas for toughest innovation challenges, trusted by R&D personnel around the world.

Constraint heteroscedasticity linear discriminant analysis method for language identification

A linear discriminant analysis and language recognition technology, applied in speech analysis, speech recognition, instruments, etc.

Inactive Publication Date: 2009-09-30
TSINGHUA UNIV
View PDF0 Cites 12 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0010] In order to overcome the problems existing when the existing HLDA algorithm is applied to high-dimensional supervectors, the present invention provides a CHLDA (Constrained Heteroskedasticity Linear Discriminant Analysis) algorithm, which greatly reduces the calculation quantity

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
  • Constraint heteroscedasticity linear discriminant analysis method for language identification
  • Constraint heteroscedasticity linear discriminant analysis method for language identification
  • Constraint heteroscedasticity linear discriminant analysis method for language identification

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0052] Assuming that c is a supervector feature, for the HLDA algorithm, its goal is to find a linear transformation matrix A such that

[0053] x=Ac

[0054] In order to obtain a new vector after transformation, the discriminative information is compressed to the first several dimensions. In the present invention, c is spliced ​​by the basic features, because DCT (discrete cosine transform) has been carried out when the basic features are obtained, the correlation in the basic feature frame is already weak, and the key programming of the problem removes the basic feature between frames relevance. Going back to the cepstral matrix C, we can only study the lateral transformation, and the corresponding transformation matrix A should have the form of a diagonal block matrix, namely

[0055]

[0056]In addition, according to the characteristics of supervector features, we can prove that its covariance matrix has the form of an approximate diagonal block matrix through theoret...

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 provides a constraint heteroscedasticity linear discriminant analysis method for language identification, which relates to a method for the dimension reduction and decorrelation of high-dimension feature vectors. The method is characterized in that MFCC features are extracted from voice signals; the MFCC features of continuous M frames are selected and placed in parallel so as to obtain a cepstrum matrix; the cepstrum matrix is expanded according to rows to form super vectors; the mean and covariance of the super vectors are calculated by the block; a transformation matrix is calculated by the block by an iteration method; the super vectors are transformed by the block by use of the transformation matrix; and each block is subjected to dimension reduction and decorrelation treatment so as to obtain new feature vectors. The method has the advantage of obtaining the feature vectors of which the correlation among dimensions is removed, along with small amount of calculation, high discriminant property and low dimension, and can be used for language identification.

Description

technical field [0001] The invention belongs to the field of speech recognition, and in particular relates to a constrained heteroskedasticity linear discriminant analysis method, which can be used for rapid dimensionality reduction and decorrelation processing of high-dimensional feature vectors in language recognition. Background technique [0002] Language recognition refers to the use of machines to identify the type of language from a speech signal. Language recognition is mainly used in systems such as multilingual man-machine spoken dialogue, speech translation, cross-language speech retrieval, and speech listening. [0003] Currently, the most commonly used feature in language recognition is MFCC (Mel Frequency Cepstral Coefficient) and its derived features. The most successful feature in the derived feature is the SDC (Shifted Difference Cepstrum) feature, which makes full use of the timing information in the speech signal and improves the discrimination of the fea...

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): G10L15/02
Inventor 张卫强刘加
Owner TSINGHUA 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
Eureka Blog
Learn More
PatSnap group products