Voice recognition apparatus, vehicle having the same, and method of controlling the vehicle

• US 9,756,161 B2
• Filed: 12/02/2014
• Issued: 09/05/2017
• Est. Priority Date: 08/22/2014
• Status: Active Grant

• Current Assignee: Hyundai Motor Company (Hyundai Motor Group), Kia Motors Corporation (Kia Corp.)
• Original Assignee: Hyundai Motor Company (Hyundai Motor Group)
• Inventors: Bang, KyuSeop
• Primary Examiner(s): Jia, Xin
• Application Number:
  US14/558,476
• Publication Number:
  US 20160057261A1
• Time in Patent Office: 1,008 Days
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 2015/027 Syllables being the recognition units
• CPC Class Codes: H04M 1/271 controlled by voice recognition
• CPC Class Codes: H04M 1/6091 including a wireless interface
• CPC Class Codes: H04M 2250/74 with voice recognition means voice activated dialling H04M1/271; voice recognition algorithms G10L15/00

Abstract

A voice recognition apparatus including: a context model created by modeling recognition target context; and a voice recognizer configured to create a candidate group corresponding to a received voice signal based on an acoustic model and the context model, and to apply a high reliability weight value to a candidate having a length of syllables corresponding to a length of speech.

Claims

1. A voice recognition apparatus comprising:

a context model created by modeling recognition target context;

a model creator configured to classify the recognition target context according to a length of syllables, and to create a context model for each length of syllables; and

a voice recognizer configured to create a candidate group corresponding to a received voice signal based on an acoustic model and the context model for each length of syllables, calculate a length of a user'"'"'s speech based on a Begin of Speech (BoS) and an End of Speech (EoS), and apply a high reliability weight value to a candidate having a length of syllables corresponding to the length of speech.

2. The voice recognition apparatus according to claim 1, further comprising a display unit configured to display the candidate group in which one or more candidates are arranged according to degrees of reliability.

3. The voice recognition apparatus according to claim 1, wherein if the length of speech is shorter than or equal to a threshold length, the voice recognizer applies a high reliability weight value to a candidate corresponding to a length of syllables that is shorter than or equal to a reference length of syllables.

4. The voice recognition apparatus according to claim 1, wherein the voice recognizer creates candidate groups for context models created for lengths of syllables, and applies different weight values to the candidate groups, respectively, according to the length of syllables corresponding to the length of speech.

5. The voice recognition apparatus according to claim 1, wherein the recognition target context is a phone book.

6. A vehicle comprising:

a context model created by modeling each name included in a phone book;

a model creator configured to classify the phone book according to a length of syllables of each name, and to create a context model for each length of syllables; and

a voice recognizer configured to create a phone number candidate group corresponding to a received voice signal based on an acoustic model and the context model for each length of syllables, calculate a length of speech based on a Begin of Speech (BoS) and an End of Speech (EoS), and to respectively apply different reliability weight values determined according to the length of speech to phone number candidates in the phone number candidate group.

7. The vehicle according to claim 6, further comprising a display unit configured to display the phone number candidate group in which one or more phone number candidates are arranged according to degrees of reliability.

8. The vehicle according to claim 6, wherein the voice recognizer applies the reliability weight values such that a phone number candidate having a length of syllables corresponding to the length of speech in the phone number candidate group has a high degree of reliability.

9. The vehicle according to claim 6, wherein if the length of speech is shorter than a threshold length, the voice recognizer applies the reliability weight values such that a phone number candidate of a monosyllable in the phone number candidate group has a high degree of reliability.

10. The vehicle according to claim 6, wherein if the length of speech is longer than a threshold length, the voice recognizer applies the reliability weight values such that a phone number candidate of a monosyllable in the phone number candidate group has a low degree of reliability.

11. The vehicle according to claim 6, wherein the model creator segments the name into a first name and a last name to create a segmented context model.

12. The vehicle according to claim 6, wherein the model creator converts the name into phonemes, and models the name in the form of a voice lattice.

13. The vehicle according to claim 6, wherein the voice recognizer creates a phone number candidate group corresponding to the voice signal from each context model, and applies a high weight value to a phone number candidate group of a length of syllables corresponding to the length of speech.

14. The vehicle according to claim 6, wherein the voice recognizer recognizes a phoneme string of the voice signal based on the acoustic model, and searches for one or more phone number candidates corresponding to the recognized phoneme string in the context model to create the phone number candidate group.

15. The vehicle according to claim 6, further comprising a length-of-speech measuring unit configured to detect a Begin of Speech (BoS) and an End of Speech (EoS) from the voice signal, and to measure the length of speech based on the BoS and the EoS.

16. A control method of a vehicle, comprising:

creating a phone number candidate group corresponding to a received voice signal from a context model created by modeling each name included in a phone book;

classifying the phone book according to lengths of names, and creating a context model for each of the lengths of the names;

calculate a length of speech based on a Begin of Speech (BoS) and an End of Speech (EoS);

determining reliability weight values according to the length of the user'"'"'s speech; and

applying the reliability weight values according to lengths of syllables of phone number candidates selected from the context model for each of the lengths of the names.

17. The control method according to claim 16, wherein the determining of the reliability weight values comprises measuring the length of speech based on a difference between a Begin of Speech (BoS) and an End of Speech (EoS).

18. The control method according to claim 16, wherein the determining of the reliability weight values comprises determining the reliability weight values such that a phone number candidate of a length of syllables corresponding to the length of speech has a high degree of reliability.

19. The control method according to claim 16, wherein the determining of the reliability weight values comprises determining the reliability weight values such that a phone number candidate of a monosyllable has a high degree of reliability if the length of speech is shorter than or equal to a threshold length.

20. The control method according to claim 16, wherein the determining of the reliability weight values comprises determining the reliability weight values such that a phone number candidate of a monosyllable has a low degree of reliability if the length of speech is longer than or equal to a threshold length.

21. The control method according to claim 16, wherein the creating of the phone number candidate group comprises:

detecting a feature vector from the voice signal; and

recognizing a phoneme string of the voice signal from an acoustic model according to the feature vector.

22. The control method according to claim 21, wherein the creating of the phone number candidate group comprises creating the phone number candidate group based on a degree of reliability with regard to the phoneme string.

23. The control method according to claim 21, wherein the phone number candidate group has a degree of reliability that is higher than or equal to predetermined criteria with regard to the phoneme string.

24. The control method according to claim 16, wherein the applying of the reliability weight values comprises:

arranging one or more phone number candidates of the phone number candidate group according to degrees of reliability; and

displaying the phone number candidate group with the phone number candidates arranged according to the degrees of reliability.

25. The control method according to claim 16, further comprising:

receiving a voice signal according to the user'"'"'s speech; and

pre-processing the voice signal.

26. The control method according to claim 16, wherein the creating of the context model comprises:

converting each name into a phoneme string; and

creating a voice lattice based on the phoneme string.