Voice recognition method and voice recognition apparatus

• US 9,196,247 B2
• Filed: 03/18/2013
• Issued: 11/24/2015
• Est. Priority Date: 04/27/2012
• Status: Active Grant

• Current Assignee: Fujitsu Limited
• Original Assignee: Fujitsu Limited
• Inventors: Harada, Shouji
• Primary Examiner(s): Godbold, Douglas
• Application Number:
  US13/846,234
• Publication Number:
  US 20130289992A1
• Time in Patent Office: 981 Days
• US Class Current: 1/1
• CPC Class Codes: G10L 15/20 Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech G10L21/02 takes precedence

Abstract

A voice recognition method includes: detecting a vocal section including a vocal sound in a voice, based on a feature value of an audio signal representing the voice; identifying a word expressed by the vocal sound in the vocal section, by matching the feature value of the audio signal of the vocal section and an acoustic model of each of a plurality of words; and selecting, with a processor, the word expressed by the vocal sound in a word section based on a comparison result between a signal characteristic of the word section and a signal characteristic of the vocal section.

Claims

1. A voice recognition method comprising:

detecting a vocal section including a vocal sound in a voice, based on a feature value of an audio signal representing the voice;

identifying a word expressed by the vocal sound in the vocal section, by matching the feature value of the audio signal of the vocal section and an acoustic model of each of a plurality of words; and

selecting, with a processor, the word expressed by the vocal sound in a word section based on a comparison result between a signal characteristic of the word section and a signal characteristic of the vocal section, wherein

the selecting includes selecting the word expressed by the vocal sound in the word section having a signal characteristic not less than a given lower limit threshold value and not greater than a given upper limit threshold value with respect to the signal characteristic of the vocal section, and

the signal characteristic of the word section and the signal characteristic of the vocal section are one of Signal-to-Noise Ratio (SNR) or Average Power.

2. The voice recognition method according to claim 1, wherein

the selecting includes selecting the word expressed by the vocal sound in the word section having a signal characteristic not less than a given lower limit threshold value with respect to the signal characteristic of the vocal section.

3. The voice recognition method according to claim 2, wherein

the selecting includes using the lower limit threshold value associated with each target word of the selecting.

4. The voice recognition method according to claim 3, further comprising:

calculating the lower limit threshold value associated with each target word based on reading information of each target word.

5. The voice recognition method according to claim 4, wherein

the calculating includes calculating the lower limit threshold value associated with each target word, by an average value of threshold values associated with individual syllables of reading of each word, respectively.

6. The voice recognition method according to claim 1, wherein

the identifying includes

obtaining a matching score indicating a height of similarity between the feature value of the audio signal of the voice section and the acoustic model of each of the plurality of words, and

changing the matching score based on the comparison result between the signal characteristic of the word section and the signal characteristic of the vocal section, and

the selecting includes selecting the word expressed by the vocal sound in the word section based on the matching score for the word.

7. The voice recognition method according to claim 6, wherein

the changing includes changing the matching score of the word expressed by the vocal sound in the word section having a signal characteristic less than a given lower limit threshold value with respect to the signal characteristic of the vocal section, so as to lower the height of the similarity expressed by the matching score.

8. The voice recognition method according to claim 1, further comprising:

calculating the signal characteristic of the vocal section based on a signal characteristic of a given section of the audio signal that includes the vocal section.

9. A voice recognition apparatus comprising:

a processor, coupled to a memory, configured to;

detect a vocal section including a vocal sound in a voice, based on a feature value of an audio signal representing the voice,

identify a word expressed by the vocal sound in the vocal section, by matching the feature value of the audio signal of the vocal section and an acoustic model of each of a plurality of words, and

select the word expressed by the vocal sound in a word section based on a comparison result between a signal characteristic of the word section and a signal characteristic of the vocal section, wherein

the processor is configured to select the word expressed by the vocal sound in the word section having a signal characteristic not less than a given lower limit threshold value and not greater than a given upper limit threshold value with respect to the signal characteristic of the vocal section, and

the signal characteristic of the word section and the signal characteristic of the vocal section are one of Signal-to-Noise Ratio (SNR) or Average Power.

10. The voice recognition apparatus according to claim 9, wherein

the processor is configured to select the word expressed by the vocal sound in the word section having a signal characteristic not less than a given lower limit threshold value with respect to the signal characteristic of the vocal section.

11. The voice recognition apparatus according to claim 10, wherein

the processor is configured to use the lower limit threshold value associated with each word to be a target of the selection process.

12. The voice recognition apparatus according to claim 11, wherein

the processor is further configured to calculate the lower limit threshold value associated with each target word based on reading information of each target word.

13. The voice recognition apparatus according to claim 12, wherein

the processor is configured to calculate the lower limit threshold value associated with each target word, by an average value of threshold values associated with individual syllables of reading of each word, respectively.

14. The voice recognition apparatus according to claim 9, wherein

the processor is further configured to:

obtain a matching score indicating a height of similarity between the feature value of the audio signal of the voice section and the acoustic model of each of the plurality of words, and

change the matching score based on the comparison result between the signal characteristic of the word section and the signal characteristic of the vocal section, and

the processor is configured to select the word expressed by the vocal sound in the word section based on the matching score for the word.

15. The voice recognition apparatus according to claim 14, wherein

the processor is configured to change the matching score of the word expressed by the vocal sound in the word section having a signal characteristic less than a given lower limit threshold value with respect to the signal characteristic of the vocal section, so as to lower the height of the similarity expressed by the matching score.

16. The voice recognition apparatus according to claim 9, herein

the processor is further configured to calculate the signal characteristic of the vocal section based on a signal characteristic of a given section of the audio signal that includes the vocal section.

17. A non-transitory computer-readable recording medium having stored therein a program for causing a computer to execute a voice recognition process comprising:

detecting a vocal section including a vocal sound in a voice, based on feature value of an audio signal representing the voice;

identifying a word expressed by the vocal sound in the vocal section, by matching the feature value of the audio signal of the vocal section and an acoustic model of each of a plurality of words; and

selecting the word expressed by the vocal sound in a word section based on a comparison result between a signal characteristic of the word section and a signal characteristic of the vocal section, wherein

the selecting includes selecting the word expressed by the vocal sound in the word section having a signal characteristic not less than a given lower limit threshold value and not greater than a given upper limit threshold value with respect to the signal characteristic of the vocal section, and

the signal characteristic of the word section and the signal characteristic of the vocal section are one of Signal-to-Noise Ratio (SNR) or Average Power.