Hybrid Echo Cancellation for Residual Echo and Fast Convergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing echo cancellation technologies suffer from inaccurate error estimation, leading to slow convergence and insufficient steady-state performance, and are computationally intensive, with no effective way to reduce complexity and improve effectiveness.
Innovation Solution
A hybrid echo cancellation method combining an acoustic echo canceller (AEC) module based on a linear filter with a residual echo cancellation (RES) module using a deep neural network (DNN) to process audio signals, effectively eliminating both linear and nonlinear echoes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional filters are used for echo cancellation, then the system is simple to implement, but the error estimation is inaccurate resulting in slow convergence and insufficient steady-state performance
Solution Approach 1:
The echo cancellation process is divided into two independent modules: linear echo cancellation (LEC) for linear echoes and residual echo cancellation (REC) for nonlinear echoes. Each module uses specialized algorithms optimized for its specific task, with LEC handling the bulk of linear echo removal and REC refining the result by addressing remaining nonlinear components. This segmentation allows each module to achieve high precision without compromising convergence speed.
Solution Approach 2:
The LEC module acts as an intermediary between the raw microphone signal and the REC module. It first removes the dominant linear echo components, producing a pre-processed signal that the REC module then further refines. This intermediate processing stage enables the REC module to focus computational resources on the more challenging nonlinear residual echoes, improving overall precision without sacrificing convergence speed.
2Measurement precision
If Kalman filters are used for echo cancellation, then the steady-state performance is improved, but the computational complexity increases
Solution Approach 1:
The cancellation system is segmented into LEC and REC modules with different computational complexities. The LEC module uses computationally efficient linear filtering algorithms to handle the majority of echo removal, while the simpler REC module uses nonlinear processing only on the residual signal. This segmentation achieves high steady-state performance without requiring the high computational complexity of a full Kalman filter implementation throughout the entire processing chain.
Solution Approach 2:
The complex Kalman filter algorithms are extracted and replaced with simpler, more efficient algorithms in the LEC module, while only the essential nonlinear processing is retained in the REC module. This extraction of complexity from the main processing path reduces overall computational requirements while maintaining the steady-state performance benefits through the specialized REC module.
3Reliability
If traditional echo cancellation methods are used, then the computational resources are consumed, but the echo cancellation effectiveness is insufficient
Solution Approach 1:
The echo cancellation task is segmented into two stages with different resource requirements. The LEC stage handles the high-volume linear echo removal with efficient algorithms, while the REC stage processes only the residual nonlinear echoes with more sophisticated but computationally lighter operations. This segmentation improves cancellation effectiveness by addressing both linear and nonlinear components while reducing overall computational resource consumption compared to applying complex algorithms to the entire signal.
Solution Approach 2:
Instead of applying full-complexity algorithms to the entire microphone signal, the system applies lightweight LEC processing to the full signal and then applies more sophisticated REC processing only to the residual echo portion. This partial application of advanced processing where most needed (to the residual echo) improves effectiveness while minimizing computational resource usage.
Data Source
AI summary
The present application provides a method for processing an audio signal, an electronic device, and a computer-readable storage medium. The present application relates to the technical field of audio processing. The method for processing the audio signal includes: obtaining a current far-end signal and a microphone signal; the microphone signal includes a near-end signal and an echo signal generated by a speaker playing the far-end signal; performing linear echo cancellation processing on the microphone signal to obtain a linear filtered signal; inputting the linear filtered signal and the far-end signal into a pre-trained residual echo cancellation DNN model to output a gain signal corresponding to the near-end signal; and determining a target audio signal to be output to the far-end for playback based on the gain signal and the linear filtered signal.