Neural Acoustic Echo Suppression for Nonlinear Full-Duplex Speech
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing acoustic echo cancellation (AEC) methods are inefficient for non-linear echo distortion, echo path changes, and non-stationary noises, particularly in full-duplex speech communication and speech recognition systems.
Innovation Solution
A NeuralEcho model and automatic gain control (AGC) are trained in a unified model using a recurrent neural network (RNN) to perform acoustic echo cancellation (AEC) and improve echo and noise suppression by leveraging second-order statistics and multi-head self-attention mechanisms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If linear adaptive filtering methods are used for acoustic echo cancellation, then the system complexity is low and robustness is good, but the method is inefficient for non-linear echo distortion, echo path changes, and non-stationary noises
Solution Approach 1:
The patent combines linear adaptive filtering methods with deep neural networks to form a hybrid system. The linear filter handles basic echo cancellation with low complexity, while the DNN component addresses non-linear echo distortion and non-stationary noises, achieving both robustness and adaptability.
Solution Approach 2:
The deep neural network is designed to perform multiple functions including non-linear echo distortion suppression, echo path change adaptation, and non-stationary noise suppression, making the system versatile for various acoustic conditions while maintaining the simplicity of linear filtering for standard cases.
2Adaptability or versatility
If deep neural networks are employed for echo suppression to handle non-linear echo distortions, then the handling capability for non-linear distortions improves, but the system complexity increases
Solution Approach 1:
The echo suppression system is segmented into multiple components: a linear adaptive filter for basic echo cancellation and a deep neural network for non-linear distortion handling. This segmentation allows each component to specialize in specific tasks, improving overall performance while managing complexity through functional division.
Data Source
AI summary
A method of acoustic echo suppression using a recurrent neural network, performed by at least one processor, is provided. The method includes receiving a microphone signal and a far-end reference signal, estimating an echo suppressed signal and an echo signal based on the microphone signal and the far-end reference signal, estimating enhancement filters for the microphone signal based on the echo suppressed signal and the echo signal, generating an enhanced signal based on the enhancement filters, and adjusting the enhanced signal using automatic gain control (AGC) and outputting the adjusted signal.