US 2014/0064476 A1 · SYSTEMS AND METHODS OF ECHO & NOISE CANCELLATION IN VOICE COMMUNICATION

Abstract

In an example, time and frequency domain speech enhancement is implemented on a platform having a programmable device, such a PC or a smartphone running an OS. Echo cancellation is done first in time domain to cancel a dominant portion of the echo. Residual echo is cancelled jointly with noise reduction during a subsequent frequency domain stage. The time domain block uses a dual band, shorter length Adaptive Filter for faster convergence. Non-linear residual echo is cancelled based on an echo estimate and an error signal from the adaptive filters. A controller locates regions that had residual echo suppressed and which do not have speech and injects comfort noise. The controller can be full-duplex and operate non-linearly. An AGC selectively amplifies the frequency bins, based on the Gain function used by the residual echo and noise canceller.

Claims33

Dependent claims

A machine-implemented method of echo cancellation in a full duplex voice communication system, comprising:

performing acoustic echo cancellation on a near-end signal using a short time domain adaptive filter, to produce a filtered near-end signal that may have non-linear residual echo and noise;
tracking, in the frequency domain, the non-linear residual echo in the filtered near-end signal to output an error signal;
producing an estimate of a portion of the filtered near-end signal to be removed, as a combination of the non-linear residual echo, and noise;
imposing a limitation on the estimate based on a moving average of the error signal;
controlling gains associated with a plurality of frequency bins in a Frequency domain Automatic Gain Controller (FAGC), based on the limited estimate, thereby suppressing the estimated portion of the filtered near-end signal to be removed;
and
refining respective gains associated with the plurality of frequency bins of the FAGC.

4 more elements

A system for controlling gain of a near-end signal in a voice enhancement system, the near-end signal being a speech signal of a user at a near end, the system comprising:

a module for identifying a plurality of frequency bins which contain speech energy;
a local gain estimator operating to estimate local gain for each of the plurality of frequency bins;
a speech identifier operable to identify bins of the plurality of frequency bins that contain speech;
a local gain smoothing module operable to smooth gain in the plurality of frequency bins;
and
a global gain regulator operable to estimate and regular a global gain over all of the plurality of frequency bins.

3 more elements

A system for controlling a Non-Linear Processor (NLP) to activate and deactivate the NLP for complete removal of residual echo in an echo alone region of a microphone output signal without chopping of a near-end speech signal, the system comprising:

an estimator configured to produce respective estimates for a plurality of decision parameters;
a detector for detecting convergence of an adaptive echo cancellation filter;
a controller to output a NLP decision for a frame of speech, wherein the NLP decision indicates whether the NLP is to be active or inactive;
a module for updating an NLP energy threshold parameter;
a Single Talk (ST) hangover breaker;
a Double Talk (DT) hangover breaker;
and
a module for revising the NLP decision.

5 more elements

Dependencies are inferred from the claim text.

Systems and methods of echo & noise cancellation in voice communication

Life of this patent

Systems and methods of echo & noise cancellation in voice communication

Life of this patent

Abstract

Claims33