Blind Subband Acoustic Echo Postfiltering for Residual Echo Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional acoustic echo cancellation methods leave residual echo that is still audible, requiring non-linear postfiltering processes which need knowledge of the far-end audio signal leaking back through microphones or audio sensors.
Innovation Solution
The implementation of a blind subband acoustic echo postfiltering system that processes the acoustic signal without knowledge of the far-end signal, using a noise suppression system and a blind subband AEC postfilter (BSAP) system to render residual echo inaudible by determining echo dominance estimates and generating combined echo/noise masks for each subband.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If conventional acoustic echo cancellation is applied, then echo is reduced, but residual echo remains audible
Solution Approach 1:
The patent divides the acoustic signal into multiple frequency subbands using a filter bank. Each subband is processed independently by the postfilter to selectively attenuate residual echo while preserving speech. This segmentation allows precise control over different frequency components where echo and speech may have different characteristics.
Solution Approach 2:
The postfilter applies different filtering characteristics to different subbands based on local signal properties. The echo dominance estimate and noise estimate are calculated separately for each subband, allowing the filter to adapt its attenuation strength locally rather than applying uniform processing across the entire frequency spectrum.
2Object-affected harmful factors
If non-linear postfiltering is applied to reduce residual echo, then echo suppression is improved, but knowledge of far-end signal is required
Solution Approach 1:
The blind postfilter operates autonomously using only the near-end microphone signal and the AEC masked signal as inputs. It estimates echo dominance and noise characteristics directly from these signals without requiring external far-end signal information. The system serves itself by deriving all necessary filtering parameters from the available local signals.
Solution Approach 2:
The patent introduces an intermediary processing stage between AEC and noise suppression that estimates echo dominance in each subband. This intermediary module bridges the gap by providing echo-related information derived from local signals, enabling the postfilter to function without direct access to far-end signal while still achieving effective echo suppression.
3Loss of information
If blind subband postfiltering is applied, then far-end signal knowledge is not required, but system complexity increases
Solution Approach 1:
The system uses a filter bank to segment the signal into subbands, enabling parallel processing of multiple frequency components. This segmentation structure organizes the complexity into manageable modules, where each subband processor handles a specific frequency range independently, making the overall complex system more tractable and efficient.
Solution Approach 2:
The postfilter applies partial attenuation to each subband based on the estimated echo dominance, rather than attempting complete echo removal. By applying just enough filtering to make residual echo inaudible while preserving speech quality, the system achieves effective echo suppression without requiring excessively complex processing for each subband.
Data Source
AI summary
Systems and methods for blind subband acoustic echo cancellation postfiltering in a communication device are provided. In exemplary embodiments, an acoustic signal is received via a microphone of the communication device. Acoustic echo cancellation (AEC) is applied to the acoustic signal to obtain an AEC masked signal. Because residual echo may still exist in the AEC masked signal, blind subband AEC postfiltering on the AEC masked signal may be performed to obtain an echo-free acoustic signal. The echo-free signal may then be output.