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

Method and device for checking voice recognition results, voice recognition system and audio monitoring system

A sound recognition and sound technology, applied in speech recognition, speech analysis, instruments, etc., can solve the problems of not considering the recognition results, sound recognition false alarms, and difficulty in setting robust thresholds, so as to improve recognition performance and reduce false alarms Effect

Active Publication Date: 2013-10-23
CANON KK
View PDF9 Cites 3 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

More specifically, in the method disclosed in prior art 1, since the confidence is calculated without normalization, it is difficult to set a robust threshold under different conditions
In the method disclosed in prior art 2, it only uses the difference of the recognition scores of the first candidate sound and the second candidate sound to determine the confidence level, but no other relationship between the recognition scores in the N-best list is used
[0014] In addition, in traditional inspection methods, the confidence level is only determined based on the current recognition results, without considering the recognition results over a longer period of time (that is, the influence of background noise)
[0015] The above issues can affect the accuracy of the confidence score and thus lead to false alarms during the sound recognition process

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
  • Method and device for checking voice recognition results, voice recognition system and audio monitoring system
  • Method and device for checking voice recognition results, voice recognition system and audio monitoring system
  • Method and device for checking voice recognition results, voice recognition system and audio monitoring system

Examples

Experimental program
Comparison scheme
Effect test

no. 1 example

[0053] figure 2 is a flowchart showing a method for checking a voice recognition result according to the first embodiment of the present invention.

[0054] Such as figure 2 As shown, in the receiving step 210, the N-best list of the voice recognition results of the current window is received from the voice recognition engine.

[0055] During the voice recognition process, voice signals are input into the voice recognition engine. Then, a series of potential recognition candidate sounds is generated with their corresponding recognition scores. The sound recognition engine sorts the recognition candidate sounds into an N-best list according to the recognition scores of the recognition candidate sounds, and then outputs the N-best list.

[0056] In the first calculation step 220, a first probability distribution of all candidate sounds in the N-best list of the current window is calculated based on the N-best list of the current window.

[0057] For each candidate sound in...

no. 2 example

[0096] Figure 5 is a flowchart showing a method for checking a voice recognition result according to the second embodiment of the present invention. as from Figure 5 As can be seen from , the difference between the method according to the second embodiment and the method according to the first embodiment is that a determination step 540 is added before the third calculation step is performed. In the sound recognition process according to the second embodiment, it is intended to recognize (detect) a target sound (or a plurality of target sounds). If the first candidate sound (which has the highest recognition score in the N-best list) of the current window's N-best list is not a target sound, it is not necessary to check this sound recognition result.

[0097] Such as Figure 5 As shown, in the receiving step 510, the N-best list of the voice recognition results of the current window is received from the voice recognition engine.

[0098] In the first calculation step 520...

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 method and a device for checking voice recognition results, a voice recognition system and an audio monitoring system used for automatically detecting abnormal audio events. The method for checking the voice recognition results comprises a receiving step, a first calculation step, a second calculation step, a third calculation step and a checking step, wherein the receiving step refers to receiving N pieces of best lists of voice recognition results of the current window from a voice recognition engine; the first calculation step refers to calculating first probability distribution of all candidate voices in the N pieces of best lists on the basis of the N pieces of best lists of the current window; the second calculation step refers to calculating second probability distribution of all the candidate voices on the basis of N pieces of best lists of a long window of the current window; the third calculation step refers to calculating the distance between the first probability distribution and the second probability distribution so as to be used as a confidence coefficient; and the checking step refers to detecting the voice recognition results of the current window by using the confidence coefficient. Thanks to the invention, false alarm can be reduced, and the recognition performance can be improved.

Description

technical field [0001] The present invention relates to a method and apparatus for verifying voice recognition results, a voice recognition system and an audio monitoring system for automatic detection of abnormal audio events. Background technique [0002] Confidence Measure (CM) technology is usually used to reduce false alarms in the process of voice recognition. More specifically, after the voice recognition result is obtained, the confidence level is calculated based on the voice recognition result. Then, the confidence is compared with a predetermined threshold, thereby checking the voice recognition result. Confidence is a score used to assess the reliability of voice recognition results. In many practical applications, a good confidence level can greatly benefit the sound recognition process. [0003] Generally, the sound recognition results are output in the form of N best (N-best) lists, and the N-best lists are composed of the N best candidate sounds sorted and...

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 Applications(China)
IPC IPC(8): G10L15/00G10L25/78
Inventor 郭莉莉沈海峰
Owner CANON KK