Transient sound event detection

• US 9,911,414 B1
• Filed: 12/20/2013
• Issued: 03/06/2018
• Est. Priority Date: 12/20/2013
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Yang, Jun
• Primary Examiner(s): Vo, Huyen
• Application Number:
  US14/136,923
• Time in Patent Office: 1,537 Days
• Field of Search: 704203, 704205-215, 704219-230, 704500-504
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 25/51 for comparison or discrimination
• CPC Class Codes: G10L 25/84 for discriminating voice from noise

Abstract

Techniques are described for detecting and/or validating audible taps, double-taps, or other tapped audio sequences generated by a user. A predicted voice signal is subtracted from an audio signal to produce a prediction error signal. The prediction error signal is analyzed in terms of dynamics and periodicity to determine whether it contains an audio pulse. In certain cases, prediction error signals corresponding to multiple microphones are analyzed to determine position information regarding the individual taps of a double-tap. A double-tap is validated and acted upon when the individual taps originate from the same location and the taps occur within a given time period.

Claims

1. A method comprising:

receiving an audio signal from a microphone;

determining a portion of the audio signal corresponding to speech;

determining a prediction error signal that excludes the portion corresponding to the speech;

analyzing the prediction error signal to detect a first audible tap that occurred from a first location and a second audible tap that occurred from a second location;

identifying a portion of the prediction error signal that corresponds in time to the first audible tap;

identifying a portion of the prediction error signal that corresponds in time to the second audible tap;

calculating a first time-difference-of-arrival (TDOA) value based at least in part on the portion of the prediction error signal corresponding in time to the first audible tap;

calculating a second TDOA value based at least in part on the portion of the prediction error signal corresponding in time to the second audible tap;

determining that the first location and the second location are a common location based at least in part on the first TDOA value and the second TDOA value;

determining that the first audible tap and the second audible tap occurred within a predefined time period;

determining an audio double-tap has occurred based at least in part on (a) the first audible tap and the second audible tap originating from the common location and (b) the first audible tap and the second audible tap occurring within the predefined time period; and

causing a computing device to execute a command based at least in part on determining that the audio double-tap has occurred.

2. The method of claim 1, wherein determining that the first location and the second location are a common location comprises:

analyzing the first TDOA value to determine the first location; and

analyzing the second TDOA value to determine the second location.

3. The method of claim 1, wherein determining that the first audible tap and the second audible tap occurred from a common location comprises calculating a distance between the first location and the second location based at least in part on the first TDOA value and the second TDOA value.

4. The method of claim 1, wherein the analyzing comprises:

determining an envelope value of the prediction error signal, wherein the envelope value comprises a smoothed amplitude boundary of the prediction error signal;

determining a floor value of the prediction error signal, wherein the floor value comprises a smoothed average amplitude of the prediction error signal;

analyzing the envelope value with respect to the floor value to determine a dynamic range;

determining a periodicity; and

determining an occurrence of at least one of the first audible tap or the second audible tap when the dynamic range is greater than a first threshold and the periodicity is less than a second threshold.

5. A method comprising:

receiving an audio signal from a microphone;

determining a portion of the audio signal corresponding to speech;

determining a prediction error signal that excludes the portion corresponding to the speech;

determining a dynamic range of the prediction error signal;

determining a periodicity of the prediction error signal;

determining, by a processor, an occurrence of an audible tap based at least in part on the dynamic range being greater than a first threshold and the periodicity being less than a second threshold; and

causing a computing device to execute a command based at least in part on determining the occurrence of the audible tap.

6. The method of claim 5, wherein determining the portion of the audio signal corresponding to the speech comprises performing linear predictive coding.

7. The method of claim 5, further comprising processing the audio signal to reduce stationary noise in the audio signal.

8. The method of claim 5, wherein determining the dynamic range comprises:

determining an envelope value of the prediction error signal, wherein the envelope value comprises a smoothed amplitude boundary of the prediction error signal;

determining a floor value of the prediction error signal, wherein the floor value comprises a smoothed average amplitude of the prediction error signal; and

analyzing the envelope value with respect to the floor value.

9. The method of claim 5, wherein determining the periodicity comprises auto-correlating signal values related to the prediction error signal.

10. The method of claim 5, wherein determining the dynamic range comprises using a first sample rate and determining the periodicity comprises using a second sample rate that is lower than the first sample rate.

11. The method of claim 5, wherein determining the occurrence of the audible tap is based at least in part on determining that the audible tap originated more than a predefined distance from the microphone.

12. The method of claim 5, further comprising:

calculating a time-difference-of-arrival (TDOA) value based at least in part on the prediction error signal; and

analyzing the TDOA value to determine an originating location of the audible tap.

13. The method of claim 5, wherein the audio signal is based on at least one of audible audio or ultrasonic audio.

14. A method comprising:

receiving an audio signal;

determining a portion of the audio signal corresponding to speech;

determining, by one or more processors, a prediction error signal via linear predictive coding, the prediction error signal excluding the portion corresponding to the speech;

determining, by the one or more processors, an occurrence of an audible tap based at least in part on analyzing the prediction error signal; and

causing a computing device to execute a command based at least in part on determining the occurrence of the audible tap.

15. The method of claim 14, wherein determining the portion of the audio signal corresponding to the speech comprises performing linear predictive coding.

16. The method of claim 14, wherein analyzing the prediction error signal comprises:

determining an envelope value of the prediction error signal, wherein the envelope value comprises a smoothed amplitude boundary of the prediction error signal;

determining a floor value of the prediction error signal, wherein the floor value comprises a smoothed average amplitude of the prediction error signal;

analyzing the envelope value with respect to the floor value to determine a dynamic range of the prediction error signal; and

determining a periodicity of the prediction error signal.

17. The method of claim 16, wherein determining the periodicity comprises auto-correlating signal values related to the prediction error signal.

18. The method of claim 16, wherein determining the occurrence of the audible tap is based at least in part on the dynamic range being greater than a first threshold and the periodicity being less than a second threshold.

19. The method of claim 14, further comprising:

processing the audio signal to produce a down-sampled audio signal;

determining a portion of the down-sampled audio signal corresponding to the speech;

determining a second prediction error signal that excludes the portion of the down-sampled audio signal corresponding to the speech; and

wherein determining the occurrence of the audible tap is further based at least in part on analyzing the second prediction error signal.

20. The method of claim 14, further comprising:

processing the audio signal to produce a down-sampled audio signal;

determining a portion of the down-sampled audio signal corresponding to the speech;

determining a second prediction error signal that excludes the portion of the down-sampled audio signal corresponding to the speech;

determining an envelope value of the second prediction error signal, wherein the envelope value comprises a smoothed amplitude boundary of the second prediction error signal;

determining a floor value of the second prediction error signal, wherein the floor value comprises a smoothed average amplitude of the second prediction error signal;

analyzing the envelope value with respect to the floor value to determine a dynamic range of the second prediction error signal;

determining a periodicity of the second prediction error signal; and

wherein determining the occurrence of the audible tap is based at least in part on determining that the dynamic range is greater than a first threshold and the periodicity is less than a second threshold.

21. The method of claim 14, further comprising:

calculating a time-difference-of-arrival (TDOA) value based at least in part on the prediction error signal; and

analyzing the TDOA value to determine an originating location of the audible tap.