ADAPTIVE AUDIO FRAME PROCESSING FOR KEYWORD DETECTION

• US 20140337030A1
• Filed: 12/10/2013
• Published: 11/13/2014
• Est. Priority Date: 05/07/2013
• Status: Abandoned Application

• Current Assignee: Qualcomm, Inc.
• Original Assignee: Qualcomm, Inc.
• Inventors: Lee, Minsub, Kim, Taesu, Hwang, Kyu Woong, Kim, Sungwoong, Jin, Minho
• Application Number:
  US14/102,097
• Publication Number:
  US 20140337030A1
• Time in Patent Office: Days
• US Class Current: 704/251
• 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 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 15/32 Multiple recognisers used in sequence or in parallel; Score combination systems therefor, e.g. voting systems

Abstract

A method of detecting a target keyword from an input sound for activating a function in a mobile device is disclosed. In this method, a first plurality of sound features is received in a buffer, and a second plurality of sound features is received in the buffer. While receiving each of the second plurality of sound features in the buffer, a first number of the sound features are processed from the buffer. The first number of the sound features includes two or more sound features. Further, the method may include determining a keyword score for each of the processed sound features and detecting the input sound as the target keyword if at least one of the keyword scores is greater than a threshold score.

Claims

1. A method of detecting a target keyword from an input sound for activating a function in a mobile device, the method comprising:

receiving a first plurality of sound features in a buffer;

receiving a second plurality of sound features in the buffer;

while receiving at least one sound feature of the second plurality of sound features in the buffer, processing a first number of sound features from the buffer, the first number of sound features including two or more sound features;

determining a keyword score for at least one sound feature of the processed sound features; and

detecting the input sound as the target keyword if the keyword score is greater than a threshold score.

2. The method of claim 1, wherein processing the first number of sound features comprises:

retrieving the first number of sound features from the buffer; and

processing the retrieved sound features based on feature statistics.

3. The method of claim 2, wherein the feature statistics include a mean and a variance, and wherein processing the retrieved sound features comprises normalizing the retrieved sound features based on the feature statistics.

4. The method of claim 2, wherein processing the retrieved sound features comprises processing a sound feature in the first plurality of sound features based on the feature statistics of the first plurality of sound features.

5. The method of claim 2, wherein processing the retrieved sound features comprises:

updating the feature statistics with a sound feature in the second plurality of sound features; and

processing the sound feature in the second plurality of sound features based on the updated feature statistics.

6. The method of claim 2, wherein processing the retrieved sound features comprises processing a selected number of the retrieved sound features based on the feature statistics.

7. The method of claim 6, wherein the selected number is adjusted based on resource information of the mobile device.

8. The method of claim 2, wherein processing the retrieved sound features comprises:

identifying a sound feature among the retrieved sound features as a current sound feature;

determining a difference between the current sound feature and a previous sound feature; and

if the difference is less than a threshold value, using a processed sound feature of the previous sound feature as a processed sound feature of the current sound feature.

9. The method of claim 1, wherein the first number is adjusted based on resource information of the mobile device.

10. The method of claim 1, wherein determining the keyword score for the at least one sound feature of the processed sound features comprises calculating the keyword score using a Markov chain model.

11. The method of claim 1, wherein processing the first number of sound features comprises, if a particular number of the sound features in the buffer is less than the first number, processing the particular number of sound features that are in the buffer.

12. The method of claim 1,

wherein receiving the first plurality of sound features in the buffer comprises:

segmenting a first portion of the input sound into a first plurality of frames; and

extracting a first sound feature from at least one frame of the first plurality of frames,

wherein receiving the second plurality of sound features in the buffer comprises;

segmenting a second portion of the input sound into a second plurality of frames; and

extracting a second sound feature from at least one frame of the second plurality of frames.

13. The method of claim 1, further comprising activating the function associated with the target keyword in response to the input sound being detected as the target keyword.

14. A mobile device, comprising:

a buffer configured to store a first plurality of sound features and a second plurality of sound features;

a feature processing unit configured to process a first number of sound features from the buffer while the buffer receives each of the second plurality of sound features, the first number of the sound features including two or more sound features;

a keyword score calculation unit configured to determine a keyword score for each of the processed sound features; and

a keyword detection unit configured to detect an input sound as a target keyword if at least one of the keyword scores is greater than a threshold score.

15. The mobile device of claim 14, wherein the feature processing unit is further configured to:

retrieve the first number of the sound features from the buffer; and

process the retrieved sound features based on feature statistics.

16. The mobile device of claim 15, wherein the feature statistics include a mean and a variance, and wherein the feature processing unit is further configured to normalize the retrieved sound features based on the feature statistics.

17. The mobile device of claim 15, wherein the feature processing unit is further configured to process a sound feature in the first plurality of sound features based on feature statistics.

18. The mobile device of claim 15, wherein the feature statistics are updated based on statistics of a sound feature in the second plurality of sound features, and wherein the feature processing unit is further configured to process the sound feature in the second plurality of sound features based on the updated feature statistics.

19. The mobile device of claim 15, wherein the feature processing unit is further configured to process a selected number of the retrieved sound features based on the feature statistics.

20. The mobile device of claim 19, wherein selected number is adjusted based on resource information of the mobile device.

21. The mobile device of claim 15, wherein the feature processing unit is further configured to:

identify a sound feature among the retrieved sound features as a current sound feature;

determine a difference between the current sound feature and a previous sound feature; and

if the difference is less than a threshold value, use a processed sound feature of the previous sound feature as a processed sound feature of the current sound feature.

22. The mobile device of claim 14, wherein the first number is adjusted based on resource information of the mobile device.

23. The mobile device of claim 14, wherein the feature processing unit is further configured to, if the number of the sound features in the buffer is less than the first number, process the sound features that are in the buffer.

24. The mobile device of claim 14, wherein the keyword detection unit is further configured to activate a function associated with the target keyword based on the input sound being detected as the target keyword.

25. A mobile device, comprising:

means for storing sound features, wherein the sound features include a first plurality of sound features and a second plurality of sound features;

means for processing a first number of sound features from the means for storing the sound features while the means for storing the sound features receives each of the second plurality of sound features, the first number of the sound features including two or more sound features;

means for determining a keyword score for each of the processed sound features; and

means for detecting an input sound as a target keyword if at least one of the keyword scores is greater than a threshold score.

26. The mobile device of claim 25, wherein the means for processing the first number of the sound features is configured to:

retrieve the first number of the sound features from the means for storing the sound features; and

process the retrieved sound features based on feature statistics.

27. The mobile device of claim 26, wherein the feature statistics include a mean and a variance, and wherein the means for processing the first number of the sound features is further configured to normalize the retrieved sound features based on the feature statistics.

28. A non-transitory computer-readable storage medium storing instructions for detecting a target keyword from an input sound for activating a function in a mobile device, the instructions causing a processor to perform operations, the operations comprising:

receiving a first plurality of sound features in a buffer;

receiving a second plurality of sound features in the buffer;

while receiving each of the second plurality of sound features in the buffer, processing a first number of the sound features from the buffer, the first number of the sound features including two or more sound features;

determining a keyword score for each of the processed sound features; and

detecting the input sound as the target keyword if at least one of the keyword scores is greater than a threshold score.

29. The non-transitory computer-readable storage medium of claim 28, wherein the first number is adjusted based on resource information of the mobile device.

30. The non-transitory computer-readable storage medium of claim 28, wherein processing the first number of sound features comprises, if a particular number of sound features in the buffer is less than the first number, processing the particular number of sound features that are in the buffer.