Granted Patent B2

US 10,504,538 · App. 15/611,499 · Granted Dec 10, 2019

Inventor: Jeffrey Bullough (South Jordan, UT)

Assignee: SORENSON IP HOLDINGS, LLC

G10L21/0232G10L25/84G10L19/012G10L19/0204G10L21/0208

Abstract

Claims (59)

1. A computer-implemented method to reduce noise in a signal, the method comprising:

obtaining an audio signal that includes speech;

separating the audio signal into frequency components in each of a plurality of frequency bands;

obtaining a first magnitude threshold that is based on one or more characteristics of human speech in a first frequency band of the plurality of frequency bands;

obtaining a second magnitude threshold that is based on one or more characteristics of human speech in a second frequency band of the plurality of frequency bands, the second magnitude threshold being different than the first magnitude threshold;

calculating a first average magnitude of first frequency components in the first frequency band and a second average magnitude of second frequency components in the second frequency band during a first time frame;

calculating a third average magnitude of the first frequency components and a fourth average magnitude of the second frequency components during a second time frame that is after the first time frame;

in response to a difference or ratio between the first average magnitude and the third average magnitude of the first frequency band being less than the first magnitude threshold, attenuating the first frequency components based on an interpolation of the ratio between the first magnitude threshold and a third magnitude threshold;

in response to a difference or ratio between the second average magnitude and the fourth average magnitude of the second frequency band being less than the second magnitude threshold, attenuating the second frequency components; and

combining the frequency components, including the attenuated first frequency components and the attenuated second frequency components, to produce an output audio signal.

2. The method of claim 1 , wherein the one or more characteristics of human speech in the first frequency band include a first range of magnitudes of one or more phonemes in the first frequency band and wherein the one or more characteristics of human speech in the second frequency band include a second range of magnitudes of one or more phonemes in the second frequency band, the one or more phonemes in the second frequency band differing from the one or more phonemes in the first frequency band.

3. The method of claim 1 , wherein each of the plurality of frequency bands includes an approximately equal bandwidth of frequency.

4. The method of claim 1 , wherein attenuating the first frequency components comprises attenuating the first frequency components by a fixed percentage amount.

5. The method of claim 1 , wherein attenuating the first frequency components comprises attenuating the first frequency components based on a difference between the first average magnitude and the second average magnitude.

6. The method of claim 1 , wherein the one or more characteristics of human speech in the first frequency band include phonemes of human speech in the first frequency band.

7. The method of claim 1 , wherein the first magnitude threshold is based on one or more characteristics of human speech in the first frequency band, a duration of the first time frame, and a duration of the second time frame.

8. The method of claim 1 , wherein the first time frame and the second time frame each comprises a duration of 50 ms.

9. The method of claim 1 , wherein the first average magnitude, the second average magnitude, the third average magnitude, and the fourth average magnitude are root mean square averages.

10. At least one non-transitory computer readable medium configured to store one or more instructions that when executed by at least one system performs the method of claim 1 .

11. A computer-implemented method to reduce noise in a signal, the method comprising:

obtaining an audio signal;

separating the audio signal into frequency components in each of a plurality of frequency bands;

obtaining a first threshold for a first frequency band of the plurality of frequency bands;

obtaining a second threshold for the first frequency band;

calculating a first envelope of first frequency components in the first frequency band during a first time frame;

calculating a second envelope of the first frequency components during a second time frame that is after the first time frame;

calculating a first signal-to-noise ratio based on a difference between a magnitude of the first envelope and a magnitude of the second envelope;

in response to the first signal-to-noise ratio being greater than the first threshold and less than the second threshold, attenuating the first frequency components by an amount based on interpolation of the first signal-to-noise ratio between the first threshold and the second threshold; and

combining the frequency components, including the attenuated first frequency components, to produce an output audio signal.

12. The method of claim 11 , wherein each of the plurality of frequency bands includes an approximately equal bandwidth of frequency.

13. The method of claim 11 , further comprising:

obtaining a third threshold for a second frequency band of the plurality of frequency bands, the third threshold being different from the first threshold;

obtaining a fourth threshold for the second frequency band, the fourth threshold being different from the third threshold;

calculating a third envelope of second frequency components in the second frequency band during the first time frame;

calculating a fourth envelope of the second frequency components during the second time frame;

calculating a second signal-to-noise ratio based on a difference between a magnitude of the third envelope and a magnitude of the fourth envelope; and

in response to the second signal-to-noise ratio being greater than the third threshold and less than the fourth threshold, attenuating the second frequency components by an amount based on interpolation of the second signal-to-noise ratio between the third threshold and the fourth threshold,

wherein combining the frequency components includes combining the frequency components, including the attenuated first frequency components and the attenuated second frequency components.

14. The method of claim 11 , further comprising:

calculating a fifth envelope of the first frequency components in the first frequency band during a third time frame after the first time frame and the second time frame;

calculating a sixth envelope of the first frequency components in the first frequency band during a fourth time frame that is after the third time frame;

calculating a third signal-to-noise ratio based on a difference between a magnitude of the fifth envelope and a magnitude of the sixth envelope;

in response to the third signal-to-noise ratio being less than the first threshold, attenuating the first frequency components by a fixed percentage amount; and

combining the frequency components, including the attenuated first frequency components, to produce a second output audio signal during the third time frame.

15. The method of claim 11 , wherein the first envelope comprises a first average magnitude of the first frequency components during the first time frame and the second envelope comprises a second average magnitude of the first frequency components during the second time frame.

16. At least one non-transitory computer readable medium configured to store one or more instructions that when executed by at least one system performs the method of claim 11 .

17. A computer-implemented method to reduce noise in a signal, the method comprising:

obtaining an audio signal that includes speech;

separating the audio signal into frequency components in each of a plurality of frequency bands;

obtaining a first noise threshold that is based on one or more characteristics of human speech in a first frequency band of the plurality of frequency bands;

obtaining a second noise threshold that is based on one or more characteristics of human speech in a second frequency band of the plurality of frequency bands, the second noise threshold being different than the first noise threshold;

calculating a first signal envelope for first frequency components in the first frequency band and a second signal envelope for second frequency components in the second frequency band for a first duration of time;

calculating a third signal envelope for the first frequency components and a fourth signal envelope for the second frequency components for a second duration of time that is longer than the first duration of time and overlaps the first duration of time;

calculating a first noise ratio for the first frequency components using the first signal envelope and the third signal envelope;

calculating a second noise ratio for the second frequency components using the second signal envelope and the fourth signal envelope;

in response to the first noise ratio being less than the first noise threshold, attenuating the first frequency components by an amount based on interpolation of the first noise ratio between the first noise threshold and a third noise threshold;

in response to the second noise ratio being less than the second noise threshold, attenuating the second frequency components; and

combining the plurality of frequency bands, including the attenuated first frequency components and the attenuated second frequency components, to produce an output audio signal.

18. The method of claim 17 , wherein the one or more characteristics of human speech in the first frequency band include one or more phonemes of human speech in the first frequency band.

Assignments (13)