An adaptive filter bank based neural network approach for time delay estimation and speech enhancement

TL;DR

This work addresses time delay estimation (TDE) in adaptive-filter-based acoustic echo cancellation (AEC) by introducing an adaptive filter bank neural network (ADF-NN) that estimates delay via $M$ filters of length $N$ with overlap $L$, producing $M\times N$ block-energy features classified to yield the delay. The estimated delay guides a residual-echo suppression pipeline employing a neural network, followed by robust denoising with the OMLSA algorithm and a spectrum-smoothed automatic gain control (AGC). Experimental results on AEC Challenge 2021 data show superior TDE accuracy and increased PESQ scores compared with WebRTC baselines, while keeping model size compact; the combination of NN-based residual suppression and OMLS A post-processing reduces distortion and improves stability, aided by sigmoid-based gain smoothing to mitigate spectral discontinuities. The approach is designed to be open-source friendly, enabling reproducible implementation and practical deployment in real-world AEC systems.

Abstract

Time delay estimation (TDE) plays a key role in acoustic echo cancellation (AEC) using adaptive filter method. Considerable residual echo will be left if estimation error arises. Here, in this paper, we proposed an adaptive filter bank based neural network approach where the delay is estimated by a bank of adaptive filters with overlapped time scope, and all the energy of filter weights are concatenated and feed to a classification network. The index with maximal probability is chosen as the estimated delay. Based on this TDE, an AEC scheme is designed using a neural network for residual echo and noise suppression, and the optimally-modified log-spectral amplitude (OMLSA) algorithm is adopted to make it robust. Also, a robust automatic gain control (AGC) scheme with spectrum smoothing method is designed to amplify speech segments. Performance evaluations reveal that higher performance can be achieved for our scheme.

Figures (5)

• Figure 1: Algorithm structure of the proposed time delay estimation and speech enhancement.
• Figure 2: Schematic of TDE training data preparation.
• Figure 3: Data preparation for suppression network.
• Figure 4: Depiction of gain discontinuity.
• Figure 5: Schematic of gain smoothing.