Combining Adaptive Filtering And Complex-Valued Deep Postfiltering For Acoustic Echo Cancellation

Abstract

In this contribution, we introduce a novel approach to noise-robust acoustic echo cancellation employing a complex-valued Deep Neural Network (DNN) for postfiltering. In a first step, early linear echo components are removed using a double-talk robust adaptive filter. The residual signal is subsequently processed by the proposed postfilter (PF). Due to its complex-valued nature, the PF allows to suppress unwanted signal components without introducing distortions to the near-end speaker. For training and evaluation, we exclusively use data from the ICASSP 2021 AEC challenge. Exploiting only a moderate amount of training data, we demonstrate the efficacy of the proposed method. Specifically, we show that the PF (i) benefits significantly from a preceding linear adaptive filter and (ii) significantly outperforms a conventional real-valued DNN-based PF.

Deep Residual Echo Suppression and Noise Reduction: A Multi-Input FCRN Approach in a Hybrid Speech Enhancement System

“…In [2] a similar idea was followed, however, only for a very small subset of the possible combinations, a rather simple feedforward neural network, and without explicit consideration of noise reduction capabilities. Furthermore, an intuitive and often seen input signal combination of a RES/NR network is the pair of enhanced signal E (k) and reference signal X (k), which delivers state of the art performance, e.g., in [4,9]. But as the reference signal does not contain any information about the room characteristics yet, could the estimated echo signal D (k) be a better choice?…”

“…In this typical two-stage arrangement for speech enhancement, the application of a DNN as second stage (RES and NR) gained increasing attention with early investigations of feed-forward networks [1,2], convolutional networks bringing further improvements more recently [3,4], some even being fully synergistic with the first stage [5], and many more. In the meantime, also fully learned deep AEC approaches were proposed, where a single network incorporates the tasks of AEC, RES, and NR, e.g., [6,7] or further investigated in [8].…”

EchoFilter: End-to-End Neural Network for Acoustic Echo Cancellation

“…Since TIMIT dataset was widely used in literatures to evaluate AEC performance, we follows the data preparation method as is referred to in [17,19], resulting in 3500 training mixtures, and 300 test mixtures. The generation for inputs and labels are illustrated in Fig.…”

“…Four categories are obtained, i.e., no voice exist at both branches (i.e., silence), only voice exists at the near-end (i.e., near-end single-talk), only voice exists at the far-end (i.e., far-end single-talk), voice exist at both ends (i.e., double-talk), acting as classification labels. The RIR in the experiments are configured as is referred to in [17,19], resulting in 7 RIRs, of which the first 6 RIRs are used to generate training mixtures and the last one is used to generate test mixtures. The hard clipping is used to simulate the power amplifier of loudspeaker, and the memoryless sigmoidal function is applied to emulate the nonlinear characteristic of loudspeaker, resulting in x nl (n) for nonlinear inputs.…”

“…To overcome this problem, numeric methods have been proposed, such as Nonlinear AEC (NLAEC) method by using a set of nonlinear basis functions for echo estimation [9][10][11] and nonlinear post-filtering method by cascading an additional nonlinear processing module for residual echo suppression [12][13][14][15]. Recently, since its great potential in speech processing tasks, neural network (NN) has been used for echo cancellation or suppression, such as NN-based postfiltering method where NN is used for residual echo suppression instead of conventional post-filters [16,17], NN-based NLAEC method where NN is used for modeling nonlinear echo [18], separation-based method where source separation with an additional information from the far-end [19] is used for AEC, NNbased adaptive filtering method where the structure of conventional linear filter is adopted for designing network [20,21].…”