Feature parameter vector grouping quantization method in speech recognition

A speech recognition and feature parameter technology, applied in the field of biometrics, can solve problems such as the inability to ensure that the algorithm converges to the global optimum, and the large amount of calculation.

Inactive Publication Date: 2009-11-18
HARBIN INST OF TECH SHENZHEN GRADUATE SCHOOL
View PDF0 Cites 4 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The LBG algorithm is essentially a modification of the K value algorithm, both of which belong to the steepest descent algorithm (Steepest Descend Algorithm). The two algorithms can only reach the local optimum, and cann

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
  • Feature parameter vector grouping quantization method in speech recognition
  • Feature parameter vector grouping quantization method in speech recognition

Examples

Experimental program
Comparison scheme
Effect test

specific Embodiment approach 1

[0011] Specific embodiment one: referring to Fig. 1, this embodiment consists of the following steps:

[0012] Step A1: Group all speech data participating in training according to their characteristics.

[0013] Step A2: Select the voice recognition information group to be processed, and set the required number N of codewords.

[0014] Step A3: extract the Mel frequency cepstral coefficients (MFCCs) of all speech signals contained in the current speech recognition information group.

[0015] Step A4: Randomly select the first C MFCCs of the current speech signal as the initial central values ​​of the N MFCC codebooks, and the initial average distortion degree D old =∞; and set a cycle improvement parameter step, the initial value is 10 6 .

[0016] Step A5: according to the nearest neighbor rule based on the Euler distance, all the feature parameters MFCC of the current speech signal are mapped to the current V codeword regions to form N regions;

[0017] Step A6: Find ea...

specific Embodiment approach 2

[0025] Specific implementation mode two: this implementation mode further illustrates the specific method of the nearest neighbor method described in step A5 on the basis of specific implementation mode one:

[0026] Step B1: Calculate the distortion measure based on the Euclidean distance, the distortion measure is defined as the square of the distance between the vector X and the corresponding feature point in the feature space, d 2 ( x , z i ) = Σ k = 1 N ( x i - ( Z i ) K ) ...

specific Embodiment approach 3

[0028] Specific implementation mode three: this implementation mode further illustrates on the basis of specific implementation mode one that the specific method for eliminating the empty cell cavity and the average number of vectors in the cell cavity described in step A6 is:

[0029] Step C1: Find each region, if some region has no input vector into this category, it is determined that this region is an empty cavity, and marked.

[0030] Step C2: Find out the area with the largest number of vectors by item-by-item comparison, find out the feature vectors belonging to this area, divide the vector into this area, and classify it into the empty cell cavity marked in the previous step until the empty cell The number of vectors in the cavity is half of the number of vectors in the largest area.

[0031] Step C3: Find the center of the re-divided area to form a new codebook.

[0032] Step C4: Skip back to C1 until there is no empty cavity.

[0033] The above algorithm steps illu...

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

A feature parameter vector grouping quantization method in speech recognition relates to a vector quantization method in speech recognition methods, belonging to the field of biological recognition. The current feature parameter vector quantization method in the speech recognition generally establishes a code book for multiple words; as speech information differences among the words are greater, and the range of the code book is limited, frames of different words can fall in a same code word region by the quantization of the code book, thus causing problems for improving the subsequent speech recognition rate. The feature parameter vector grouping quantization method leads all speech training information of all words to constitute a speech group, the code book after the vector quantization has great relevance to the information of the corresponding group, feature frames containing the same information in the speech information of different people for the same word can fall in the same code word region with very large probability, thereby providing a basis for the follow-up accurate calculation of various parameters.

Description

technical field [0001] The invention relates to a speech recognition method, in particular to a vector quantization method in the speech recognition method, and belongs to the field of biological recognition. Background technique [0002] Speech recognition belongs to a kind of biometric technology, which is a technology for machines to convert human voice signals into corresponding text or commands through the process of recognition and understanding. Speech recognition mainly includes three modules: parameter description, model approximation, and inference judgment. After the speech signal feature parameters are extracted, it is an important step in speech recognition parameter description to carry out reasonable and effective vector quantization coding on the feature parameters. [0003] The vector quantization coding process mainly includes two parts: [0004] Optimal division: Under the premise of a given number of vector quantizations, all codebooks are mapped to eac...

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): G10L19/02G10L15/06
Inventor 王明江余芳张爽李硕刘德
Owner HARBIN INST OF TECH SHENZHEN GRADUATE SCHOOL
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