Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Model output post-processing method and deviceand computer equipment

A model output, computer program technology, applied in speech analysis, speech recognition, instruments, etc., can solve the problems of long calculation time and large amount of post-processing algorithm calculation, so as to reduce the process of repeated calculation, ensure numerical stability, reduce The effect of calculation

Active Publication Date: 2021-12-07
深圳市友杰智新科技有限公司
View PDF22 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] The main purpose of this application is to provide a post-processing method, device and computer equipment for model output, aiming to solve the disadvantages of the existing speech recognition post-processing algorithm, which has a large amount of calculation and takes too long to calculate

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
  • Model output post-processing method and deviceand computer equipment
  • Model output post-processing method and deviceand computer equipment
  • Model output post-processing method and deviceand computer equipment

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0020] In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application, and are not intended to limit the present application.

[0021] refer to figure 1 , an embodiment of the present application provides a post-processing method for model output, including:

[0022] S1: Obtain the probability matrix output by the speech recognition model;

[0023] S2: Perform a window search in the probability matrix, during the window search process, perform regular processing on the probability matrix based on a threshold, remove invalid paths in the probability matrix, and the next window interval of the probability matrix The probability calculation is based on the probability results of the previous window ...

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 model output post-processing method and device and computer equipment. A recognition system obtains a probability matrix output by a speech recognition model, carries out window search in the probability matrix, carries out rule processing on the probability matrix based on a threshold value, eliminates invalid paths in the probability matrix, the probability calculation of the next window interval of the probability matrix is calculated based on the probability result of the previous window interval, and the probability value corresponding to each window interval is calculated. Finally, the command word corresponding to the window interval with the highest probability value is selected as a speech recognition result. According to the method and device, the numerical stability of the probability matrix is effectively ensured through threshold-based rule processing. Log domain calculation is replaced by a probability accumulation calculation mode, so that the calculation amount is reduced. In the window searching process, the probability calculation of the latter window interval is based on the probability result of the former window interval, so that the repeated calculation process is reduced, the calculation amount is further reduced, and the processing speed is effectively accelerated.

Description

technical field [0001] The present application relates to the technical field of speech recognition, in particular to a post-processing method, device and computer equipment for model output. Background technique [0002] The existing post-processing algorithm of speech recognition is based on the probability matrix output by the neural network, and searches for the optimal path on the basis of the probability matrix. The search for the optimal path is usually obtained by using the probability multiplication algorithm, but due to the consideration of numerical stability, it is generally processed in the form of logsum (first convert the probability value into a log value, and convert the multiplication between the probability values ​​into a log value Addition), and then converted to the original probability value after the calculation is completed. [0003] Although this processing method can bring the benefits of numerical stability, due to the introduction of a large num...

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): G10L15/22
CPCG10L15/22G10L2015/223
Inventor 李杰王广新杨汉丹
Owner 深圳市友杰智新科技有限公司
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com