NN3A: Neural Network supported Acoustic Echo Cancellation, Noise Suppression and Automatic Gain Control for Real-Time Communications

TL;DR

This paper introduces NN3A, a neural network-based algorithm that enhances real-time communication quality by integrating acoustic echo cancellation, noise suppression, and automatic gain control into a unified model, outperforming existing methods.

Contribution

The paper presents a novel multi-task neural network model for RTC that combines residual echo suppression, noise reduction, and speech activity detection, with a new loss function to balance residual suppression and speech quality.

Findings

Outperforms separate and end-to-end models in RTC tasks.

Identifies a trade-off between echo suppression and speech distortion.

Introduces a novel loss weighting function to optimize model performance.

Abstract

Acoustic echo cancellation (AEC), noise suppression (NS) and automatic gain control (AGC) are three often required modules for real-time communications (RTC). This paper proposes a neural network supported algorithm for RTC, namely NN3A, which incorporates an adaptive filter and a multi-task model for residual echo suppression, noise reduction and near-end speech activity detection. The proposed algorithm is shown to outperform both a method using separate models and an end-to-end alternative. It is further shown that there exists a trade-off in the model between residual suppression and near-end speech distortion, which could be balanced by a novel loss weighting function. Several practical aspects of training the joint model are also investigated to push its performance to limit.