Dynamically adjusting a voice recognition system

Patent metadata

Field	Value
Publication number	US-9984688-B2
Application number	US-201615278632-A
Country	US
Kind code	B2
Filing date	Sep 28, 2016
Priority date	Sep 28, 2016
Publication date	May 29, 2018
Grant date	May 29, 2018

Abstract

A system for dynamically adjusting a voice recognition system is provided herein. The system employs a methodology for determining a display associated with a disambiguation window, with the determination employing a detection system scoring to display disambiguation based on the vocal clarity of the speaker and other environmental factors that may affect detection.

First claim

We claim:

1. A system for dynamically adjusting a voice recognition system, comprising:
a data store comprising a non-transitory computer readable medium storing a program of instructions for the implementation of the sorted linked list;
a processor that executes the program of instructions, the instruction comprising the following steps:
receiving a voice command from the voice recognition system;
retrieving a plurality of detected words based on the voice command;
retrieving a plurality of scores correlated to each of the plurality of detected words; and
selectively displaying a disambiguation window containing a subset of the plurality of detected words based on a first method;

wherein the first method comprises:
performing a check for disambiguation, in response to the plurality of detected words only consisting of one word, bypassing the displaying of the disambiguation window; and
in response to the bypassing of the displaying, executing a command based on the one word associated with the plurality of detected words;

wherein the performing the check for disambiguation comprises the following steps:
disregarding a set of the plurality of detected words below a predetermined score;
determining whether the non-disregarded set of the plurality of detected words is greater than one, and
in response to the determination being false, bypassing the displaying of the disambiguation window;
in response to the determination being true,
performing a second method to determine the selection threshold;
obtaining a difference between a first highest scored one of the plurality of detected words and a second highest scored one of the plurality of detected words,
in response to the difference being greater than the determined selection threshold, bypassing the displaying of the disambiguation window; and
in response to difference being greater than the determined selection threshold, displaying the disambiguation window with the plurality of detected words with a respective score within the determined selection threshold,

wherein the second method comprises:
receiving a lookup table with a plurality of predetermined trigger words;
determining whether the highest scored one of the plurality of detected words matches with multiple entries from the lookup table, with the matching being defined as one of the detected words being the same as one of the predetermined trigger words by at least a matching predetermined threshold;
in response to the determination of whether the highest scored one of the plurality of detected words matches with multiple entries from the lookup table being true, determining the selection threshold value to be greater than a default selection threshold value, and

in response to the determination of whether the highest scored one of the plurality of detected words matches with multiple entries from the lookup table being false, determining the selection threshold value is a default selection threshold value.

2. A system for dynamically adjusting a voice recognition system, comprising:
a data store comprising a non-transitory computer readable medium storing a program of instructions for the implementation of the sorted linked list;
a processor that executes the program of instructions, the instruction comprising the following steps:
receiving a voice command from the voice recognition system;
retrieving a plurality of detected words based on the voice command;
retrieving a plurality of scores correlated to each of the plurality of detected words; and
selectively displaying a disambiguation window containing a subset of the plurality of detected words based on a first method;

wherein the first method comprises:
performing a check for disambiguation, in response to the plurality of detected words only consisting of one word, bypassing the displaying of the disambiguation window; and
in response to the bypassing of the displaying, executing a command based on the one word associated with the plurality of detected words;

wherein the performing the check for disambiguation comprises the following steps:
disregarding a set of the plurality of detected words below a predetermined score;
determining whether the non-disregarded set of the plurality of detected words is greater than one, and
in response to the determination being false, bypassing the displaying of the disambiguation window;
in response to the determination being true,
performing a second method to determine the selection threshold;
obtaining a difference between a first highest scored one of the plurality of detected words and a second highest scored one of the plurality of detected words,
in response to the difference being greater than the determined selection threshold, bypassing the displaying of the disambiguation window; and
in response to difference being greater than the determined selection threshold, displaying the disambiguation window with the plurality of detected words with a respective score within the determined selection threshold,

wherein the second method comprises:
receiving a list of predetermined trigger words;
determining whether the plurality of detected words is above a predetermined number;
in response to the determination of whether the plurality of detected words is above a predetermined number being false, determining the selection threshold is a default selection threshold;
in response to the determination of whether the plurality of detected words is above a predetermined number being true,
determining whether any of the predetermined number of highest scored of the plurality of detected words match any of the list of predetermined trigger words by a matching score over a predetermined threshold,
in response to being true, determining the selection threshold is a default selection threshold; and
in response to being false, increasing the determined selection threshold by a predetermined amount.

Assignees

• Lester Brian
• Panainte Sorin M
• Visteon Global Tech Inc

Inventors

• Lester Brian
• Panainte Sorin M

Classifications

• G10L15/22 Primary

  Procedures used during a speech recognition process, e.g. man-machine dialogue · CPC title

• G10L15/10

  using distance or distortion measures between unknown speech and reference templates · CPC title

• G10L2015/225

  Feedback of the input speech · CPC title

• G10L2015/088

  Word spotting · CPC title

• G10L2015/223

  Execution procedure of a spoken command · CPC title