Linear scoring for low power wake on voice

• US 10,170,115 B2
• Filed: 07/12/2018
• Issued: 01/01/2019
• Est. Priority Date: 12/23/2016
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Bocklet, Tobias, Dorau, Tomasz, Sobon, Przemyslaw, Tomaszewski, Przemyslaw
• Primary Examiner(s): Shah, Paras D
• Assistant Examiner(s): Le, Thuykhanh
• Application Number:
  US16/034,006
• Publication Number:
  US 20180322876A1
• Time in Patent Office: 173 Days
• Field of Search: None
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 15/16 using artificial neural networks
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 2015/088 Word spotting
• CPC Class Codes: G10L 2015/223 Execution procedure of a spoken command

Abstract

Key phrase detection techniques for applications such as wake on voice are discussed include performing a vectorized operation on a multiple element acoustic score vector for a current time instance including a single state rejection model score and scores for a multiple state key phrase model and a multiple element state score vector for a previous time instance including a previous state score for the single state rejection model and previous state scores for the multiple state key phrase model to generate a multiple element score summation vector and a second vectorized operation on the multiple element score summation vector to determine a multiple element state score vector for the current time instance. The multiple element state score vector for the current time instance may then be evaluated to determine whether received audio input includes a key phrase corresponding to the multiple state key phrase model.

Claims

1. A system to perform key phrase detection comprising:

memory to store a state score vector and an acoustic score vector; and

a processor coupled to the memory, the processor to;

combine the state score vector and the acoustic score vector to generate a score summation vector, wherein the state score vector comprises, for a first time instance, a state score for a single state rejection model and state scores for a multiple state key phrase model corresponding to a predetermined key phrase, and wherein the acoustic score vector comprises, for a second time instance and based on received audio input, an acoustic score for the single state rejection model and acoustic scores for the multiple state key phrase model;

determine an updated state score vector for the second time instance based on the multiple element score summation vector, wherein the updated state score vector comprises an updated state score for the single state rejection model and updated state scores for the multiple state key phrase model;

evaluate the updated state score for the single state rejection model and a selected updated state score for the multiple state key phrase model to detect the predetermined key phrase in the received audio input;

provide at least one of a system wake indicator or a system command in response to the detected predetermined key phrase.

2. The system of claim 1, wherein the score summation vector comprises a rejection state value corresponding to a sum of the acoustic score for the single state rejection model and the state score for the single state rejection model.

3. The system of claim 2, wherein the score summation vector comprises key phrase model values corresponding to sums of the acoustic scores for the multiple state key phrase model and the state scores for the multiple state key phrase model.

4. The system of claim 3, wherein the processor to determine the updated state score vector comprises the processor to determine a maximum of the rejection state value and a first value of the key phrase model values.

5. The system of claim 4, wherein the processor to determine the updated state score vector comprises the processor to determine a maximum of the first value and a second value of the key phrase model values.

6. The system of claim 5, wherein the processor to determine the maximum of the rejection state value and the first value of the key phrase model values and the processor to determine the maximum of the first value and the second value of the key phrase model values are to be performed simultaneously.

7. The system of claim 4, wherein the first value corresponds to an initial state of the multiple state key phrase model.

8. The system of claim 4, wherein the processor to determine the updated state score vector comprises the processor to determine maxima between adjacent remaining values of the key phrase model values to provide the updated state scores for the multiple state key phrase model.

9. The system of claim 1, wherein the processor to combine the state score vector and the acoustic score vector is to add corresponding elements of the state score vector and the acoustic score vector simultaneously.

10. The system of claim 1, the processor to generate the acoustic score vector by being configured to:

determine the score for the single state rejection model as a maximum of a best rejection score corresponding to the single state rejection model and a best silence score corresponding to the single state rejection model; and

access a deep neural network acoustic model to determine the scores for the multiple state key phrase model.

11. The system of claim 1, wherein the memory and the processor are implemented as a system on a chip.

12. A computer-implemented method for key phrase detection comprising:

combining a state score vector and an acoustic score vector to generate a score summation vector, wherein the state score vector comprises, for a first time instance, a state score for a single state rejection model and state scores for a multiple state key phrase model corresponding to a predetermined key phrase, and wherein the acoustic score vector comprises, for a second time instance and based on received audio input, an acoustic score for the single state rejection model and acoustic scores for the multiple state key phrase model;

determining an updated state score vector for the second time instance based on the multiple element score summation vector, wherein the updated state score vector comprises an updated state score for the single state rejection model and updated state scores for the multiple state key phrase model;

evaluating the updated state score for the single state rejection model and a selected updated state score for the multiple state key phrase model to detect the predetermined key phrase in the received audio input;

providing at least one of a system wake indicator or a system command in response to the detected predetermined key phrase.

13. The method of claim 12, wherein the score summation vector comprises a rejection state value corresponding to a sum of the acoustic score for the single state rejection model and the state score for the single state rejection model and key phrase model values corresponding to sums of the acoustic scores for the multiple state key phrase model and the state scores for the multiple state key phrase model.

14. The method of claim 13, wherein determining the updated state score vector comprises:

determining a maximum of the rejection state value and a first value of the key phrase model values; and

determining a maximum of the first value and a second value of the key phrase model values.

15. The method of claim 13, wherein determining the updated state score vector comprises:

determining maxima between adjacent remaining values of the key phrase model values to provide the updated state scores for the multiple state key phrase model.

16. At least one non-transitory computer readable medium comprising a plurality of instruction that, in response to being executed on a device, cause the device to perform key phrase detection by:

combining a state score vector and an acoustic score vector to generate a score summation vector, wherein the state score vector comprises, for a first time instance, a state score for a single state rejection model and state scores for a multiple state key phrase model corresponding to a predetermined key phrase, and wherein the acoustic score vector comprises, for a second time instance and based on received audio input, an acoustic score for the single state rejection model and acoustic scores for the multiple state key phrase model;

determining an updated state score vector for the second time instance based on the multiple element score summation vector, wherein the updated state score vector comprises an updated state score for the single state rejection model and updated state scores for the multiple state key phrase model;

evaluating the updated state score for the single state rejection model and a selected updated state score for the multiple state key phrase model to detect the predetermined key phrase in the received audio input;

providing at least one of a system wake indicator or a system command in response to the detected predetermined key phrase.

17. The non-transitory computer readable medium of claim 16, wherein the score summation vector comprises a rejection state value corresponding to a sum of the acoustic score for the single state rejection model and the state score for the single state rejection model and key phrase model values corresponding to sums of the acoustic scores for the multiple state key phrase model and the state scores for the multiple state key phrase model.

18. The non-transitory computer readable medium of claim 17, wherein determining the updated state score vector comprises:

determining a maximum of the rejection state value and a first value of the key phrase model values; and

determining a maximum of the first value and a second value of the key phrase model values.

19. The non-transitory computer readable medium of claim 17, wherein determining the updated state score vector comprises:

determining maxima between adjacent remaining values of the key phrase model values to provide the updated state scores for the multiple state key phrase model.

20. The non-transitory computer readable medium of claim 16, wherein generating the acoustic score vector comprises:

determining the score for the single state rejection model as a maximum of a best rejection score corresponding to the single state rejection model and a best silence score corresponding to the single state rejection model; and

accessing a deep neural network acoustic model to determine the scores for the multiple state key phrase model.