Method of correcting dictionary, program for correcting dictionary, voice processing apparatus, and robot

Patent metadata

Field	Value
Publication number	US-10636415-B2
Application number	US-201715791447-A
Country	US
Kind code	B2
Filing date	Oct 24, 2017
Priority date	Oct 31, 2016
Publication date	Apr 28, 2020
Grant date	Apr 28, 2020

Abstract

A method is provided for correcting a dictionary used in a voice processing apparatus. The method includes first extracting a speech of a target speaker from audio collected by a microphone, and estimating a speech phonemic sequence configuring the speech. The method also includes calculating a match degree, using a first dictionary, between the speech phonemic sequence and a first phonemic sequence that corresponds to a first word registered in the first dictionary, and second extracting the first word corresponding to a highest match degree as a spoken word spoken by the target speaker. The method further includes first correcting a second dictionary based on the highest match degree, the second dictionary indicating a relation between a second word and a third word, and second correcting the second dictionary by correcting the relation between the third word matching the spoken word and the second word.

First claim

What is claimed is:

1. A method of correcting a dictionary used in a voice processing apparatus, the method comprising:
first extracting a speech of a target speaker from audio collected by a microphone corresponding to the voice processing apparatus;
estimating a speech phonemic sequence configuring the speech;
calculating a match degree, using a first dictionary, between the speech phonemic sequence and a first phonemic sequence that corresponds to a first word registered in the first dictionary, the first dictionary being stored in a memory included in the voice processing apparatus and indicating a corresponding relation between the first word and the first phonemic sequence;
second extracting, from the first dictionary, the first word corresponding to a highest match degree as a spoken word spoken by the target speaker;
first correcting a second dictionary based on the highest match degree, the second dictionary indicating a corresponding relation between a second word and a third word, the third word corresponding to a language unit spoken by the target speaker intending to speak the second word; and
second correcting the second dictionary by correcting the corresponding relation between the third word matching the spoken word and the second word that corresponds to the third word.

2. The method according to claim 1 ,
wherein when the spoken word is included in the second dictionary and the highest match degree is at least a first threshold value, the second dictionary is corrected by deleting, from the second dictionary, a combination of the second word matching the spoken word and the third word that corresponds to the second word.

3. The method according to claim 1 ,
wherein the second dictionary further includes the corresponding relation between the second word and frequency of the target speaker speaking the third word which matches the spoken word, and
wherein when the spoken word is registered in the second dictionary as the second word, the highest match degree is at least the first threshold value, and the frequency of the third word that corresponds to the second word which matches the spoken word is at or less than a second threshold value, the second dictionary is corrected by deleting, from the second dictionary, the combination of the second word matching the spoken word and the third word that corresponds to the second word.

4. The method according to claim 1 ,
wherein the target speaker is a target child who is an object of voice recognition in the voice processing apparatus,
wherein whether or not the speaker is the target child is determined by comparing a voiceprint extracted from the speech and the voiceprint of the target child previously recorded in the memory, and
wherein the speech phonemic sequence is estimated when the speaker is determined to be the target child.

5. The method according to claim 1 ,
wherein the memory further stores a history table indicating the corresponding relation between the first word and the match degree, and
wherein when a process of recording the first word corresponding to the highest match degree in association with the match degree in the history table is repeated a predetermined number of times, the second dictionary is corrected, based on the match degree in the history table, by correcting the corresponding relation between the second word and the third word registered in the second dictionary.

6. A non-transitory computer readable storage medium encoded with an executable computer program for correcting a dictionary used in a voice processing apparatus, that when executed by a processor causes the processor to perform operations, the operations comprising:
first extracting a speech of a target speaker from audio collected from a microphone corresponding to the voice processing apparatus;
estimating a speech phonemic sequence configuring the speech;
calculating a match degree, using a first dictionary, between the speech phonemic sequence and a first phonemic sequence that corresponds to a first word registered in the first dictionary, the first dictionary being stored in a memory included in the voice processing apparatus and indicating a corresponding relation between the first word and the first phonemic sequence;
second extracting, from the first dictionary, the first word corresponding to a highest match degree as a spoken word spoken by the target speaker;
first correcting a second dictionary based on the highest match degree, the second dictionary indicating a corresponding relation between a second word and a third word, the third word corresponding to a language unit spoken by the target speaker intending to speak the second word; and
second correcting the second dictionary by correcting the corresponding relation between the third word matching the spoken word and second word that corresponds to the third word.

7. A voice processing apparatus, comprising:
a processor; and
a memory,
wherein the processor executes a computer program to perform operations, the operations including
first extracting a speech of a target speaker from audio collected from a microphone corresponding to the voice processing apparatus;
estimating a speech phonemic sequence configuring the speech;
calculating a match degree, using a first dictionary, between the speech phonemic sequence and a first phonemic sequence that corresponds to a first word registered in the first dictionary, the first dictionary being stored in a memory and indicating a corresponding relation between the first word and the first phonemic sequence;
second extracting, from the first dictionary, the first word corresponding to a highest match degree as a spoken word spoken by the target speaker;
first correcting a second dictionary based on the highest match degree, the second dictionary indicating the corresponding relation between a second word and a third word, the third word corresponding to a language unit spoken by the target speaker intending to speak the second word; and
second correcting the second dictionary by correcting the corresponding relation between the third word matching the spoken word and the second word that corresponds to the third word.

8. A robot, comprising:
a voice processing apparatus according to claim 7 ;
a casing in which the voice processing apparatus is incorporated; and
a displacement mechanism displacing the casing.

Assignees

• Panasonic Ip Man Co Ltd

Inventors

• Kunitake Yuji
• MIYAZAKI Ryouta
• Higuchi Seiya
• Ota Yusaku

Classifications

• G10L15/063 Primary

  Training · CPC title

• G10L15/187 Primary

  Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams · CPC title

• G10L2015/0633

  using lexical or orthographic knowledge sources · CPC title

• G10L15/02 Primary

  Feature extraction for speech recognition; Selection of recognition unit · CPC title

• G10L2015/025

  Phonemes, fenemes or fenones being the recognition units · CPC title