Linear Acoustic Echo Cancellation Research Articles

Erratum to “End-to-End Deep Learning-Based Adaptation Control for Linear Acoustic Echo Cancellation”

Erratum to “End-to-End Deep Learning-Based Adaptation Control for Linear Acoustic Echo Cancellation”

End-to-End Deep Learning-Based Adaptation Control for Linear Acoustic Echo Cancellation

End-to-End Deep Learning-Based Adaptation Control for Linear Acoustic Echo Cancellation

Nonlinear acoustic echo cancellation with kernelized adaptive filters

A well-known problem in speech communication is the occurrence of acoustic echo in hands-free telephony. There are several classical adaptive filters that have been used as a stand alone approach for linear acoustic echo cancellation(AEC). A common assumption in most AEC techniques is that the echo path is linear. However, in real scenarios, owing to distortions introduced by other acoustic artefacts in handheld devices, the echo path is nonlinear. Therefore, there is a need to employ a nonlinear echo canceller. In this paper, a kernel expansion on a family of adaptive filtering algorithms based on least mean square filter is proposed. Kernel methods help model the nonlinear echo path and attain global minimum easily by finding the optimal set of filter weights. The simulations are carried out for speech signal with and without noise under different signal-to-noise ratio values. The results show that the proposed method achieves a significant improvement in a nonlinear acoustic echo cancellation environment in terms of echo return loss enhancement.

The NLMS Algorithm with Time-Variant Optimum Stepsize Derived from a Bayesian Network Perspective

In this letter, we derive a new stepsize adaptation for the normalized least mean square algorithm (NLMS) by describing the task of linear acoustic echo cancellation from a Bayesian network perspective. Similar to the well-known Kalman filter equations, we model the acoustic wave propagation from the loudspeaker to the microphone by a latent state vector and define a linear observation equation (to model the relation between the state vector and the observation) as well as a linear process equation (to model the temporal progress of the state vector). Based on additional assumptions on the statistics of the random variables in observation and process equation, we apply the expectation-maximization (EM) algorithm to derive an NLMS-like filter adaptation. By exploiting the conditional independence rules for Bayesian networks, we reveal that the resulting EM-NLMS algorithm has a stepsize update equivalent to the optimal-stepsize calculation proposed by Yamamoto and Kitayama in 1982, which has been adopted in many textbooks. As main difference, the instantaneous stepsize value is estimated in the M step of the EM algorithm (instead of being approximated by artificially extending the acoustic echo path). The EM-NLMS algorithm is experimentally verified for synthesized scenarios with both, white noise and male speech as input signal.