Adaption-based reduction of echo and noise

Patent metadata

Field	Value
Publication number	US-9595997-B1
Application number	US-201313733033-A
Country	US
Kind code	B1
Filing date	Jan 2, 2013
Priority date	Jan 2, 2013
Publication date	Mar 14, 2017
Grant date	Mar 14, 2017

Abstract

The systems, devices, and processes described herein may generate a signal, such as an audio signal or an electrical signal, that may include echo and/or noise. The echo may be at least partially canceled based at least in part on acoustic echo cancellation or local echo cancellation, while the noise may be at least partially reduced based at least in part on noise reduction. A residual amount of echo and/or noise may then be determined. The residual echo and/or noise may be suppressed based at least in part on adaptive filtering of the signal, which may include modifying weights of an algorithm associated with an adaptive filter.

First claim

What is claimed is:

1. A system comprising:
memory;
one or more processors;
one or more computer-executable instructions stored in the memory and
executable by the one or more processors to:
perform spatial filtering on an audio signal generated by multiple microphones,
the audio signal being associated with noise and an acoustic echo;
reduce at least a portion of the acoustic echo associated with the audio signal and reduce at least a portion of the noise associated with the audio signal to generate an output signal;
determine residual acoustic echo associated with the output signal, the residual acoustic echo representing acoustic echo that has not been removed from the audio signal;
determine residual noise within the output signal, the residual noise representing noise that has not been removed from the audio signal; and
reduce the residual acoustic echo or the residual noise associated with the output signal by:
synchronizing the audio signal and the output signal by delaying the audio signal to account for an amount of time to reduce the residual acoustic echo or the residual noise from the output signal, wherein delaying the audio signal results in a delayed audio signal;
removing, by an adaptive filter, the acoustic echo and the noise to generate an adapted signal;
determining an error signal based at least in part on a difference between the delayed audio signal and the adapted signal;
determining, based at least in part on the error signal, one or more weight modifications to the adaptive filter; and
reducing, by a second filter, at least one of the residual acoustic echo or the residual noise from the output signal based at least in part on respective values of the one or more weight modifications.

2. The system as recited in claim 1 , wherein the one or more computer-executable instructions are further executable by the one or more processors to reduce the at least one of the residual acoustic echo or the residual noise from the output signal based at least in part on dynamically modifying one or more weights of an algorithm associated with the adaptive filter to generate modified one or more weights, the modified one or more weights of the algorithm being used to identify a voice represented within a subsequent audio signal.

3. The system as recited in claim 2 , wherein the one or more computer-executable instructions are further executable by the one or more processors to:
detect speech represented within the audio signal; and
in response to detecting the speech represented within the audio signal, determine one or more words or phrases associated with the speech.

4. The system of claim 1 , wherein the system is associated with a voice-controlled device that includes a speaker, and wherein the acoustic echo is based at least in part on sound that is output by the speaker.

5. A method comprising:
under control of one or more processors of a computing device,
generating an audio signal from sound, the audio signal being associated with an acoustic echo and noise;
reducing at least a portion of the acoustic echo associated with the audio signal and reducing at least a portion of the noise associated with the audio signal to generate an output signal;
determining that the output signal is associated with one or more of residual acoustic echo or residual noise after the reducing of the acoustic echo and the reducing of the noise from the audio signal; and
reducing the residual acoustic echo or the residual noise from the output signal by:
synchronizing the audio signal and the output signal by delaying the audio signal to account for an amount of time to reduce the residual acoustic echo or the residual noise from the output signal, wherein delaying the audio signal results in a delayed audio signal;
removing, by an adaptive filter, the acoustic echo and the noise to generate an adapted signal;
determining an error signal based at least in part on a difference between the delayed audio signal and the adapted signal;
determining, based at least in part on the error signal, one or more weight modifications to the adaptive filter; and
reducing, by a second filter, at least one of the residual acoustic echo or the residual noise from the output signal based at least in part on respective values of the one or more weight modifications.

6. The method as recited in claim 5 , wherein the acoustic echo is based at least in part on a microphone capturing the sound and a speaker subsequently outputting the sound.

7. The method as recited in claim 5 , further comprising:
determining that a voice is represented within the audio signal based at least in part on an algorithm associated with the adaptive filter;
in response to determining that the voice is represented within the audio signal,
modifying one or more weights of the algorithm to generate one or more modified weights; and
transmitting respective values of the one or more modified weights to the second filter, the second filter reducing the at least one of the residual acoustic echo or the residual noise based at least in part on the respective values of the one or more modified weights.

8. The method as recited in claim 7 , further comprising:
determining a subsequent audio signal; and
modifying the one or more modified weights based at least in part on the subsequent audio signal and in response to transmitting the respective values of the one or more modified weights to the second filter.

9. The method as recited in claim 5 , further comprising:
determining that a voice is represented within the audio signal; and
in response to determining that the voice is represented within the audio signal, bypassing the audio signal with respect to the adaptive filter.

10. The method as recited in claim 9 , further comprising:
determining that one or more of the acoustic echo or the noise is included within the audio signal; and
in response to determining that one or more of the acoustic echo or the noise is included within the audio signal, reducing the acoustic echo or the noise associated with the audio signal.

11. The method as recited in claim 5 , further comprising:
detecting speech represented within the audio signal; and
in response to detecting the speech represented within the audio signal, determining one or more words or phrases of the speech.

12. A system comprising:
memory;
one or more processors;
one or more computer-executable instructions stored in the memory and executable by the one or more processors to:
reduce acoustic echo associated with an audio signal and reduce noise associated with the audio signal to generate an output signal;
determine residual acoustic echo associated with the output signal;
determine residual noise associated with the output signal; and
reduce at least one of the residual acoustic echo or the residual noise from the output signal by:
synchronizing the audio signal and the output signal by delaying the audio signal to account for an amount of time to reduce the residual acoustic echo or the residual noise from the output signal, wherein delaying the audio signal results in a delayed audio signal;
removing, by an adaptive filter, the acoustic echo and the noise to generate an adapted signal;
determining an error signal based at least in part on a difference between the delayed audio signal and the adapted signal;
determining, based at least in part on the error signal, one or more weight modifications to the adaptive filter; and
reducing, by a second filter, at least one of the residual acoustic echo or the residual noise from the output signal based at least in part on respective values of the one or more weight

Assignees

• Amazon Tech Inc

Inventors

• Yang Jun

Classifications

• G10L21/0208

  Noise filtering · CPC title

• H04B3/23 Primary

  using a replica of transmitted signal in the time domain, e.g. echo cancellers · CPC title

• G10L2021/02166

  Microphone arrays; Beamforming · CPC title

• G10L2021/02082

  the noise being echo, reverberation of the speech · CPC title

• H04M9/082

  using echo cancellers (echo cancellers per se H04B3/23) · CPC title