Generation of wake-up words

• US 9,373,321 B2
• Filed: 12/02/2013
• Issued: 06/21/2016
• Est. Priority Date: 12/02/2013
• Status: Active Grant

• Current Assignee: Cypress Semiconductor Corporation (Infineon Technologies AG)
• Original Assignee: Cypress Semiconductor Corporation (Infineon Technologies AG)
• Inventors: Bapat, Ojas Ashok, Kumatani, Kenichi
• Primary Examiner(s): Adesanya, Olujimi
• Application Number:
  US14/093,703
• Publication Number:
  US 20150154953A1
• Time in Patent Office: 932 Days
• Field of Search: 704231-257, 704/275
• US Class Current: 1/1
• CPC Class Codes: G10L 15/06 Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice G10L15/14 takes precedence
• CPC Class Codes: G10L 2015/0638 Interactive procedures
• CPC Class Codes: G10L 2015/088 Word spotting

Abstract

A method, system and tangible computer readable medium for generating one or more wake-up words are provided. For example, the method can include receiving a text representation of the one or more wake-up words. A strength of the text representation of the one or more wake-up words can be determined based on one or more static measures. The method can also include receiving an audio representation of the one or more wake-up words. A strength of the audio representation of the one or more wake-up words can be determined based on one or more dynamic measures. Feedback on the one or more wake-up words is provided (e.g., to an end user) based on the strengths of the text and audio representations.

Claims

1. A method for generating one or more wake-up words, the method comprising:

receiving, using a keyboard, a text representation of the one or more wake-up words;

determining a strength of the text representation of the one or more wake-up words based on one or more static measures, wherein the determining the strength of the text representation comprises applying a Kullback-Leibler (KL) divergence calculation between the one or more wake-up words and words unrelated to the one or more wake-up words;

comparing a result of the KL divergence calculation to a predetermined distance score associated with a decoding accuracy of a speech recognizer;

receiving, using a microphone, an audio representation of the one or more wake-up words;

determining a strength of the audio representation of the one or more wake-up words based on one or more dynamic measures; and

providing a message on a display device, wherein the message comprises one or more improvements to a likelihood that the speech recognizer recognizes the one or more wake-up words based on the strengths of the text and audio representations.

2. The method of claim 1, wherein the determining the strength of the text representation of the one or more wake-up words comprises generating a first acoustic model for the text representation of the one or more wake-up words and a second acoustic model for one or more words unrelated to the one or more wake-up words.

3. The method of claim 1, wherein the determining the strength of the text representation of the one or more wake-up words comprises determining a number of syllables in the one or more wake-up words or determining a number of phonologically different groups of phonemes associated with the one or more wake-up words.

4. The method of claim 1, wherein the applying the Kullback-Leibler (KL) divergence calculation comprises calculating a difference between a first probability distribution function associated with one or more phonemes of the one or more wake-up words and a second probability distribution function associated with the words unrelated to the one or more wake-up words.

5. The method of claim 1, wherein the determining the strength of the audio representation of the one or more wake-up words comprises performing a voice activity detection on the audio representation of the one or more wake-up words, calculating one or more confidence scores for the one or more wake-up words or a combination thereof.

17. The method of claim 1, wherein the one or more improvements comprises at least one of a number of syllables in the one or more wake-up words, usage of different sounding words, proximity of a user to the microphone when uttering the audio representation, or a combination thereof.

18. The method of claim 1, further comprising:

accepting the one or more wake-up words when the one or more wake-up words meet a static criteria associated with the text representation and a dynamic criteria associated with the audio representation.

19. The method of claim 1, further comprising:

rejecting the one or more wake-up words when the one or more wake-up words do not meet a static criteria associated with the text representation or a dynamic criteria associated with the audio representation.

20. The method of claim 1, wherein the determining the strength of the text representation comprises:

calculating a weighted average of the one or more static measures based on the one or more wake up words;

comparing the weighted average of the one or more static measures to a static criteria associated with the text representation.

6. A system for generating one or more wake-up words, the system comprising:

an interface device configured to receive a text representation of the one more wake-up words and an audio representation of the one or more wake-up words;

a wake-up-word (WUW) processing engine configured to determine a strength of the text representation of the one or more wake-up words based on one or more static measures and to determine a strength of the audio representation of the one or more wake-up words based on one or more dynamic measures, wherein to determine the strength of the text representation, the WUW processing engine is configured to apply a Kullback-Leibler (KL) divergence calculation between the one or more wake-up words and words unrelated to the one or more wake-up words and compare a result of the KL divergence calculation to a predetermined distance score associated with a decoding accuracy of a speck recognizer; and

a display device configured to provide a message on one or more improvements to a likelihood that the speech recognizer recognizes the one or more wake-up words based on the strengths of the text and audio representations.

7. The system of claim 6, wherein the interface device comprises a keyboard and a microphone.

8. The system of claim 6, wherein the WUW processing engine is configured to generate a first acoustic model for the text representation of the one or more wake-up words and a second acoustic model for one or more words unrelated to the one or more wake-up words.

9. The system of claim 6, wherein the WUW processing engine is configured to determine a number of syllables in the one or more wake-up words or determine a number of phonologically different groups of phonemes associated with the one or more wake-up words.

10. The system of claim 6, wherein the WUW processing engine is configured to calculate a difference between a first probability distribution function associated with one or more phonemes of the one or more wake-up words and a second probability distribution function associated with the words unrelated to the one or more wake-up words for the Kullback-Leibler (KL) divergence calculation.

11. The system of claim 6, wherein the WUW processing engine is configured to perform a voice activity detection on the audio representation of the one or more wake-up words, calculate one or more confidence scores for the one or more wake-up words or a combination thereof.

12. A non-transitory computer readable medium having stored therein one or more sequences of one or more instructions for execution by one or more processors to perform a method for generating one or more wake-up words, the method comprising:

receiving, using a keyboard, a text representation of the one or more wake-up words;

determining a strength of the text representation of the one or more wake-up words based on one or more static measures, wherein the determining comprises applying a Kullback-Leibler (KL) divergence calculation between the one or more wake-up words and words unrelated to the one or more wake-up words;

comparing a result to the KL divergence calculation to a predetermined distance score associated with a decoding accuracy of a speech recognizer;

receiving, using a microphone, an audio representation of the one or more wake-up words;

determining a strength of the audio representation of the one or more wake-up words based on one or more dynamic measures; and

providing a message on a display device, wherein the message comprises one or more improvements to a likelihood that the speech recognizer recognizes the one or more wake-up words based on the strengths of the text and audio representations.

13. The non-transitory computer readable medium of claim 12, wherein the determining the strength of the text representation of the one or more wake-up words comprises generating a first acoustic model for the text representation of the one or more wake-up words and a second acoustic model for one or more words unrelated to the one or more wake-up words.

14. The non-transitory computer readable medium of claim 12, wherein the determining the strength of the text representation of the one or more wake-up words comprises determining a number of syllables in the one or more wake-up words or determining a number of phonologically different groups of phonemes associated with the one or more wake-up words.

15. The non-transitory computer readable medium of claim 12, wherein the applying the Kullback-Leibler (KL) divergence calculation comprises calculating a difference between a first probability distribution function associated with one or more phonemes of the one or more wake-up words and a second probability distribution function associated with the words unrelated to the one or more wake-up words.

16. The non-transitory computer readable medium of claim 12, wherein the determining the strength of the audio representation of the one or more wake-up words comprises performing a voice activity detection on the audio representation of the one or more wake-up words, calculating one or more confidence scores for the one or more wake-up words or a combination thereof.