Acoustic echo cancellation and automatic speech recognition with random noise
AssigneeAmazon Technologies, Inc.
InventorsRobert Ayrapetian, Bjorn Hoffmeister
Life of this patent
Estimating the term…
Abstract
Features are disclosed for performing acoustic echo cancellation using random noise. The output may be used to perform speech recognition. Random noise may be introduced into a reference signal path and into a microphone signal path. The random noise introduced into the microphone signal path may be transformed based on an estimated echo path and then combined with microphone output. The random noise introduced into the reference signal path may be combined with a reference signal and then transformed. In some embodiments, the random noise in the reference signal path may be used in the absence of another reference signal, allowing the acoustic echo canceler to be continuously trained.
Claims23
A device comprising:
- a microphone configured to receive audio and produce an input signal;
- a speaker;
- a random noise generator;
- a memory storing a set of filter coefficients corresponding to a direct path between the speaker and the microphone;
- and
- one or more modules configured to:
- receive an output signal and generate audio using the output signal and the speaker;
- generate a reference signal by adding a first random noise signal to the output signal, wherein the first random noise signal is received from the random noise generator;
- receive an input signal from the microphone, wherein the input signal corresponds to the audio;
- generate a filtered noise signal by filtering a second random noise signal with the set of filter coefficients, wherein the second random noise signal is received from the random noise generator;
- generate a processed input signal by adding the input signal to the filtered noise signal;
- perform acoustic echo cancellation, using the reference signal and the processed input signal as inputs, to generate an echo-reduced signal, wherein performing the acoustic echo cancellation comprises updating a set of acoustic echo cancellation coefficients;
- and
- provide the echo-reduced signal to an automatic speech recognition service.
A method comprising:
- receiving a first signal comprising random noise;
- receiving a second signal comprising filtered random noise, wherein the filtered random noise was created using a filter that corresponds to a direct path between a microphone and a speaker;
- and
- performing acoustic echo cancellation, using the first signal and the second signal as inputs, to generate updated acoustic echo cancellation coefficients.
A device comprising:
- a noise generator;
- and
- an acoustic echo canceller;
- wherein the noise generator generates a first signal comprising random noise,
- and wherein the acoustic echo canceller:
- receives the first signal;
- receives a second signal comprising filtered random noise, wherein the filtered random noise was created using a filter that corresponds to a direct path between a microphone and a speaker;
- and
- performs acoustic echo cancellation, using the first signal and the second signal as inputs, to generate updated acoustic echo cancellation coefficients.
Dependencies are inferred from the claim text.