Joint Noise and Echo Suppression for Two-Way Audio
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional audio enhancement methods for two-way communication struggle with residual echo and non-stationary noise, particularly in full-duplex applications, due to high computational complexity and limitations in adapting to changing acoustic environments.
Innovation Solution
An integrated approach for joint noise and echo suppression is implemented, using a frequency domain acoustic echo canceller combined with a perceptually-motivated technique that incorporates prior knowledge from physics and psychoacoustics, involving a machine learning model to restore spectral envelopes and periodicity, and employing a multidelay block frequency-domain adaptive filter with learning rate control and a proportionate normalized least mean square algorithm.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If deep learning-based audio enhancement methods are used, then speech enhancement performance is improved, but computational complexity increases
Solution Approach 1:
The patent segments the audio processing task into separate functional modules: acoustic echo cancellation (AEC) for echo removal, and a residual echo suppressor (RES) for remaining echo and noise suppression. This segmentation allows each module to be optimized independently, using traditional signal processing for AEC and deep learning for RES, thereby reducing overall computational complexity while maintaining enhancement performance
Solution Approach 2:
The patent merges traditional acoustic echo cancellation with deep learning-based residual echo suppression into a unified system. The AEC module handles the bulk of echo removal using computationally efficient traditional methods, while the RES module applies deep learning only to the residual components, combining the benefits of both approaches without fully implementing computationally intensive deep learning throughout the entire processing chain
2Device complexity
If traditional spectral subtraction methods are used, then computational complexity is reduced, but echo and noise suppression effectiveness deteriorates
Solution Approach 1:
The patent changes the processing domain from the time domain (traditional spectral subtraction) to the frequency domain using Short-Time Fourier Transform (STFT). This parameter change enables better separation of echo and noise components through frequency analysis, improving suppression effectiveness while maintaining acceptable computational complexity through efficient frequency-domain algorithms
3Productivity
If full-duplex communication is implemented, then communication efficiency is improved, but echo and noise interference increases
Solution Approach 1:
The patent implements feedback mechanisms in the acoustic echo cancellation system, where the estimated echo path is continuously updated based on the residual error between expected and actual microphone signals. This adaptive feedback allows the system to dynamically adjust to changing acoustic environments in full-duplex communication, effectively suppressing echo and noise interference while maintaining communication efficiency
Data Source
AI summary
Joint noise and echo suppression may be performed for enhancing two-way audio communications. Audio data is captured at a communication device and audio data transmitted to the communication device from another communication device are used as input features to a trained machine learning model that uses the transmitted audio data as a reference signal to eliminate residual echo in the captured audio data when also suppressing noise in the captured audio data.