Energy-efficient, accelerometer-based hotword detection to launch a voice-control system

• US 10,347,249 B2
• Filed: 04/27/2017
• Issued: 07/09/2019
• Est. Priority Date: 05/02/2016
• Status: Active Grant

• Current Assignee: Regents of the University of California (University of California)
• Original Assignee: Regents of the University of California (University of California)
• Inventors: Mohapatra, Prasant, Zhang, Li, Pathak, Parth H., Wu, Muchen
• Primary Examiner(s): Colucci, Michael C
• Application Number:
  US15/498,835
• Publication Number:
  US 20170316779A1
• Time in Patent Office: 803 Days
• Field of Search: 704275, 7042701, 704257, 704251, 704226, 715825, 382103, 381 74
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 15/24 Speech recognition using non-acoustical features

Abstract

The disclosed embodiments relate to the design of a system that uses an accelerometer in a mobile device to detect hotwords, which activate a voice interface to recognize subsequent voice input. During operation, the system gathers samples comprising readings from the accelerometer in the mobile device. Next, the system calculates features from the gathered samples. The system then classifies the calculated features using a classifier, which has been trained to detect hotwords based on calculated features. Finally, if the classifier detects one or more hotwords, the system causes the mobile device to launch a voice-control system that recognizes subsequent voice input received from an audio microphone.

Claims

1. A method for using an accelerometer in a mobile device to detect hotwords, which activate a voice interface to recognize subsequent voice input, the method comprising:

gathering samples comprising readings from the accelerometer in the mobile device;

calculating features from the gathered samples;

classifying the calculated features using a classifier, which has been trained to detect hotwords based on calculated features; and

if the classifier detects one or more hotwords, causing the mobile device to launch a voice-control system that recognizes subsequent voice input received from an audio microphone.

2. The method of claim 1, wherein gathering the samples from the accelerometer additionally involves dynamically buffering the samples.

3. The method of claim 1, wherein prior to calculating the features, the method further comprises using a high-pass filter to filter out user-mobility-related interference from the gathered samples.

4. The method of claim 1, wherein prior to gathering the samples, the method further comprises training the classifier by:

generating exemplary hotword signatures based on calculated features obtained from exemplary spoken hotwords; and

using the exemplary hotword signatures to train the classifier to recognize hotwords in real time.

5. The method of claim 1, wherein the subsequent voice input comprises voice commands.

6. The method of claim 1, wherein causing the mobile device to launch the voice-control system comprises:

powering up the audio microphone; and

powering up and/or initializing the voice-control system.

7. The method of claim 6, wherein after the voice-control system finishes recognizing the subsequent voice input, the method further comprises:

powering down the audio microphone; and

deactivating and/or powering down the voice-control system.

8. The method of claim 1, wherein the accelerometer comprises a microelectromechanical systems (MEMS) accelerometer.

9. The method of claim 1, wherein the portable device comprises one of the following:

a smartphone;

a smartwatch;

a running watch;

an activity tracker;

a smart speaker; and

smartglasses.

10. A system that uses an accelerometer in a mobile device to detect hotwords, which activate a voice interface to recognize subsequent voice input, comprising:

an input, which receives samples comprising readings gathered from the accelerometer in the mobile device; and

a processing mechanism that;

calculates features from the gathered samples; and

classifies the calculated features using a classifier, which has been trained to detect hotwords based on calculated features; and

wherein if the classifier detects one or more hotwords, the voice interface activates a voice-control system to recognize subsequent voice input received from an audio microphone.

11. The system of claim 10, wherein the input includes a dynamic buffer, which used to dynamically buffer the gathered samples.

12. The system of claim 10, wherein the input includes a high-pass filter, which filters out user-mobility-related interference from the gathered samples before the features are calculated.

13. The system of claim 10, further comprising a training mechanism, which trains the classifier by:

generating exemplary hotword signatures based on calculated features obtained from exemplary spoken hotwords; and

using the exemplary hotword signatures to train the classifier to recognize hotwords in real time.

14. The system of claim 10, wherein the subsequent voice input comprises voice commands.

15. The system of claim 10, wherein while causing the mobile device to launch the voice-control system, the voice interface:

powers up the audio microphone; and

powers up and/or initializes the voice-control system.

16. The system of claim 15, wherein after the voice-control system finishes recognizing the subsequent voice input, the voice interface:

powers down the audio microphone; and

deactivates and/or powers down the voice-control system.

17. The system of claim 10, wherein the accelerometer comprises a microelectromechanical systems (MEMS) accelerometer.

18. The system of claim 10, wherein the portable device comprises one of the following:

a smartphone;

a smartwatch;

a running watch;

an activity tracker;

a smart speaker; and

smartglasses.

19. A method for using a gyroscope in a mobile device to detect hotwords, which activate a voice interface to recognize subsequent voice input, the method comprising:

gathering samples comprising readings from the gyroscope in the mobile device;

calculating features from the gathered samples;

classifying the calculated features using a classifier, which has been trained to detect hotwords based on calculated features; and

if the classifier detects one or more hotwords, causing the mobile device to launch a voice-control system that recognizes subsequent voice input received from an audio microphone.

20. The method of claim 19, wherein gathering the samples from the gyroscope additionally involves dynamically buffering the samples.

21. The method of claim 19, wherein prior to calculating the features, the method further comprises using a high-pass filter to filter out user-mobility-related interference from the gathered samples.

22. The method of claim 19, wherein prior to gathering the samples, the method further comprises training the classifier by:

generating exemplary hotword signatures based on calculated features obtained from exemplary spoken hotwords; and

using the exemplary hotword signatures to train the classifier to recognize hotwords in real time.

23. The method of claim 19, wherein the subsequent voice input comprises voice commands.

24. The method of claim 19, wherein causing the mobile device to launch the voice-control system comprises:

powering up the audio microphone; and

powering up and/or initializing the voice-control system.

25. The method of claim 24, wherein after the voice-control system finishes recognizing the subsequent voice input, the method further comprises:

powering down the audio microphone; and

deactivating and/or powering down the voice-control system.

26. The method of claim 19, wherein the gyroscope comprises a microelectromechanical systems (MEMS) gyroscope.

27. The method of claim 19, wherein the portable device comprises one of the following:

a smartphone;

a smartwatch;

a running watch;

an activity tracker;

a smart speaker; and

smartglasses.